]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
c475ae67 | 2 | * Copyright (c) 2009, 2010 Nicira Networks. |
43253595 | 3 | * Copyright (c) 2010 Jean Tourrilhes - HP-Labs. |
064af421 | 4 | * |
a14bc59f BP |
5 | * Licensed under the Apache License, Version 2.0 (the "License"); |
6 | * you may not use this file except in compliance with the License. | |
7 | * You may obtain a copy of the License at: | |
064af421 | 8 | * |
a14bc59f BP |
9 | * http://www.apache.org/licenses/LICENSE-2.0 |
10 | * | |
11 | * Unless required by applicable law or agreed to in writing, software | |
12 | * distributed under the License is distributed on an "AS IS" BASIS, | |
13 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
14 | * See the License for the specific language governing permissions and | |
15 | * limitations under the License. | |
064af421 BP |
16 | */ |
17 | ||
18 | #include <config.h> | |
19 | #include "ofproto.h" | |
20 | #include <errno.h> | |
21 | #include <inttypes.h> | |
9d82ec47 | 22 | #include <sys/socket.h> |
064af421 BP |
23 | #include <net/if.h> |
24 | #include <netinet/in.h> | |
25 | #include <stdbool.h> | |
26 | #include <stdlib.h> | |
10a24935 | 27 | #include "byte-order.h" |
064af421 BP |
28 | #include "classifier.h" |
29 | #include "coverage.h" | |
30 | #include "discovery.h" | |
31 | #include "dpif.h" | |
4f2cad2c | 32 | #include "dynamic-string.h" |
064af421 | 33 | #include "fail-open.h" |
ca0f572c BP |
34 | #include "hash.h" |
35 | #include "hmap.h" | |
064af421 BP |
36 | #include "in-band.h" |
37 | #include "mac-learning.h" | |
38 | #include "netdev.h" | |
39 | #include "netflow.h" | |
09246b99 | 40 | #include "nx-match.h" |
064af421 BP |
41 | #include "odp-util.h" |
42 | #include "ofp-print.h" | |
fa37b408 | 43 | #include "ofp-util.h" |
72b06300 | 44 | #include "ofproto-sflow.h" |
064af421 BP |
45 | #include "ofpbuf.h" |
46 | #include "openflow/nicira-ext.h" | |
47 | #include "openflow/openflow.h" | |
064af421 BP |
48 | #include "openvswitch/datapath-protocol.h" |
49 | #include "packets.h" | |
50 | #include "pinsched.h" | |
51 | #include "pktbuf.h" | |
52 | #include "poll-loop.h" | |
064af421 BP |
53 | #include "rconn.h" |
54 | #include "shash.h" | |
55 | #include "status.h" | |
fe55ad15 | 56 | #include "stream-ssl.h" |
064af421 BP |
57 | #include "svec.h" |
58 | #include "tag.h" | |
59 | #include "timeval.h" | |
4f2cad2c | 60 | #include "unixctl.h" |
064af421 | 61 | #include "vconn.h" |
5136ce49 | 62 | #include "vlog.h" |
064af421 | 63 | |
d98e6007 | 64 | VLOG_DEFINE_THIS_MODULE(ofproto); |
064af421 | 65 | |
72b06300 | 66 | #include "sflow_api.h" |
064af421 | 67 | |
064af421 | 68 | struct ofport { |
ca0f572c | 69 | struct hmap_node hmap_node; /* In struct ofproto's "ports" hmap. */ |
064af421 BP |
70 | struct netdev *netdev; |
71 | struct ofp_phy_port opp; /* In host byte order. */ | |
ca0f572c | 72 | uint16_t odp_port; |
064af421 BP |
73 | }; |
74 | ||
75 | static void ofport_free(struct ofport *); | |
76 | static void hton_ofp_phy_port(struct ofp_phy_port *); | |
77 | ||
78 | static int xlate_actions(const union ofp_action *in, size_t n_in, | |
ae412e7d | 79 | const struct flow *, struct ofproto *, |
064af421 BP |
80 | const struct ofpbuf *packet, |
81 | struct odp_actions *out, tag_type *tags, | |
6a07af36 | 82 | bool *may_set_up_flow, uint16_t *nf_output_iface); |
064af421 | 83 | |
bcf84111 | 84 | /* An OpenFlow flow. */ |
064af421 | 85 | struct rule { |
0c43ad9a | 86 | long long int used; /* Time last used; time created if not used. */ |
064af421 | 87 | long long int created; /* Creation time. */ |
064af421 | 88 | |
bcf84111 | 89 | /* These statistics: |
064af421 | 90 | * |
bcf84111 BP |
91 | * - Do include packets and bytes from facets that have been deleted or |
92 | * whose own statistics have been folded into the rule. | |
79eee1eb | 93 | * |
bcf84111 BP |
94 | * - Do include packets and bytes sent "by hand" that were accounted to |
95 | * the rule without any facet being involved (this is a rare corner | |
96 | * case in rule_execute()). | |
064af421 | 97 | * |
bcf84111 BP |
98 | * - Do not include packet or bytes that can be obtained from any facet's |
99 | * packet_count or byte_count member or that can be obtained from the | |
100 | * datapath by, e.g., dpif_flow_get() for any facet. | |
101 | */ | |
102 | uint64_t packet_count; /* Number of packets received. */ | |
103 | uint64_t byte_count; /* Number of bytes received. */ | |
104 | ||
105 | ovs_be64 flow_cookie; /* Controller-issued identifier. */ | |
106 | ||
107 | struct cls_rule cr; /* In owning ofproto's classifier. */ | |
108 | uint16_t idle_timeout; /* In seconds from time of last use. */ | |
109 | uint16_t hard_timeout; /* In seconds from time of creation. */ | |
110 | bool send_flow_removed; /* Send a flow removed message? */ | |
111 | int n_actions; /* Number of elements in actions[]. */ | |
112 | union ofp_action *actions; /* OpenFlow actions. */ | |
113 | struct list facets; /* List of "struct facet"s. */ | |
064af421 BP |
114 | }; |
115 | ||
bcf84111 BP |
116 | static struct rule *rule_from_cls_rule(const struct cls_rule *); |
117 | static bool rule_is_hidden(const struct rule *); | |
064af421 | 118 | |
bcf84111 | 119 | static struct rule *rule_create(const struct cls_rule *, |
0193b2af | 120 | const union ofp_action *, size_t n_actions, |
ca069229 | 121 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 122 | ovs_be64 flow_cookie, bool send_flow_removed); |
064af421 | 123 | static void rule_destroy(struct ofproto *, struct rule *); |
bcf84111 BP |
124 | static void rule_free(struct rule *); |
125 | ||
126 | static struct rule *rule_lookup(struct ofproto *, const struct flow *); | |
afe75089 | 127 | static void rule_insert(struct ofproto *, struct rule *); |
064af421 | 128 | static void rule_remove(struct ofproto *, struct rule *); |
bcf84111 BP |
129 | |
130 | static void rule_send_removed(struct ofproto *, struct rule *, uint8_t reason); | |
131 | ||
132 | /* An exact-match instantiation of an OpenFlow flow. */ | |
133 | struct facet { | |
134 | long long int used; /* Time last used; time created if not used. */ | |
135 | ||
136 | /* These statistics: | |
137 | * | |
138 | * - Do include packets and bytes sent "by hand", e.g. with | |
139 | * dpif_execute(). | |
140 | * | |
141 | * - Do include packets and bytes that were obtained from the datapath | |
142 | * when a flow was deleted (e.g. dpif_flow_del()) or when its | |
143 | * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS). | |
144 | * | |
145 | * - Do not include any packets or bytes that can currently be obtained | |
146 | * from the datapath by, e.g., dpif_flow_get(). | |
147 | */ | |
148 | uint64_t packet_count; /* Number of packets received. */ | |
149 | uint64_t byte_count; /* Number of bytes received. */ | |
150 | ||
151 | /* Number of bytes passed to account_cb. This may include bytes that can | |
152 | * currently obtained from the datapath (thus, it can be greater than | |
153 | * byte_count). */ | |
154 | uint64_t accounted_bytes; | |
155 | ||
156 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
157 | struct list list_node; /* In owning rule's 'facets' list. */ | |
158 | struct rule *rule; /* Owning rule. */ | |
159 | struct flow flow; /* Exact-match flow. */ | |
160 | bool installed; /* Installed in datapath? */ | |
161 | bool may_install; /* True ordinarily; false if actions must | |
162 | * be reassessed for every packet. */ | |
163 | int n_actions; /* Number of elements in actions[]. */ | |
164 | union odp_action *actions; /* Datapath actions. */ | |
165 | tag_type tags; /* Tags (set only by hooks). */ | |
166 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
167 | }; | |
168 | ||
169 | static struct facet *facet_create(struct ofproto *, struct rule *, | |
170 | const struct flow *, | |
171 | const struct ofpbuf *packet); | |
172 | static void facet_remove(struct ofproto *, struct facet *); | |
173 | static void facet_free(struct facet *); | |
174 | ||
175 | static struct facet *facet_lookup_valid(struct ofproto *, const struct flow *); | |
176 | static bool facet_revalidate(struct ofproto *, struct facet *); | |
177 | ||
178 | static void facet_install(struct ofproto *, struct facet *, bool zero_stats); | |
179 | static void facet_uninstall(struct ofproto *, struct facet *); | |
d530fcd2 | 180 | static void facet_flush_stats(struct ofproto *, struct facet *); |
bcf84111 | 181 | |
7f7ae89d BP |
182 | static void facet_make_actions(struct ofproto *, struct facet *, |
183 | const struct ofpbuf *packet); | |
bcf84111 BP |
184 | static void facet_update_stats(struct ofproto *, struct facet *, |
185 | const struct odp_flow_stats *); | |
064af421 | 186 | |
76ce9432 BP |
187 | /* ofproto supports two kinds of OpenFlow connections: |
188 | * | |
5899143f BP |
189 | * - "Primary" connections to ordinary OpenFlow controllers. ofproto |
190 | * maintains persistent connections to these controllers and by default | |
191 | * sends them asynchronous messages such as packet-ins. | |
76ce9432 | 192 | * |
5899143f | 193 | * - "Service" connections, e.g. from ovs-ofctl. When these connections |
76ce9432 BP |
194 | * drop, it is the other side's responsibility to reconnect them if |
195 | * necessary. ofproto does not send them asynchronous messages by default. | |
7d674866 BP |
196 | * |
197 | * Currently, active (tcp, ssl, unix) connections are always "primary" | |
198 | * connections and passive (ptcp, pssl, punix) connections are always "service" | |
199 | * connections. There is no inherent reason for this, but it reflects the | |
200 | * common case. | |
76ce9432 BP |
201 | */ |
202 | enum ofconn_type { | |
5899143f BP |
203 | OFCONN_PRIMARY, /* An ordinary OpenFlow controller. */ |
204 | OFCONN_SERVICE /* A service connection, e.g. "ovs-ofctl". */ | |
76ce9432 | 205 | }; |
064af421 | 206 | |
7d674866 BP |
207 | /* A listener for incoming OpenFlow "service" connections. */ |
208 | struct ofservice { | |
209 | struct hmap_node node; /* In struct ofproto's "services" hmap. */ | |
210 | struct pvconn *pvconn; /* OpenFlow connection listener. */ | |
211 | ||
212 | /* These are not used by ofservice directly. They are settings for | |
213 | * accepted "struct ofconn"s from the pvconn. */ | |
214 | int probe_interval; /* Max idle time before probing, in seconds. */ | |
215 | int rate_limit; /* Max packet-in rate in packets per second. */ | |
216 | int burst_limit; /* Limit on accumulating packet credits. */ | |
217 | }; | |
218 | ||
219 | static struct ofservice *ofservice_lookup(struct ofproto *, | |
220 | const char *target); | |
221 | static int ofservice_create(struct ofproto *, | |
222 | const struct ofproto_controller *); | |
223 | static void ofservice_reconfigure(struct ofservice *, | |
224 | const struct ofproto_controller *); | |
225 | static void ofservice_destroy(struct ofproto *, struct ofservice *); | |
226 | ||
76ce9432 BP |
227 | /* An OpenFlow connection. */ |
228 | struct ofconn { | |
229 | struct ofproto *ofproto; /* The ofproto that owns this connection. */ | |
230 | struct list node; /* In struct ofproto's "all_conns" list. */ | |
231 | struct rconn *rconn; /* OpenFlow connection. */ | |
232 | enum ofconn_type type; /* Type. */ | |
b70eac89 | 233 | int flow_format; /* One of NXFF_*. */ |
76ce9432 BP |
234 | |
235 | /* OFPT_PACKET_IN related data. */ | |
236 | struct rconn_packet_counter *packet_in_counter; /* # queued on 'rconn'. */ | |
237 | struct pinsched *schedulers[2]; /* Indexed by reason code; see below. */ | |
238 | struct pktbuf *pktbuf; /* OpenFlow packet buffers. */ | |
239 | int miss_send_len; /* Bytes to send of buffered packets. */ | |
240 | ||
241 | /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow | |
242 | * requests, and the maximum number before we stop reading OpenFlow | |
243 | * requests. */ | |
064af421 BP |
244 | #define OFCONN_REPLY_MAX 100 |
245 | struct rconn_packet_counter *reply_counter; | |
76ce9432 | 246 | |
5899143f | 247 | /* type == OFCONN_PRIMARY only. */ |
9deba63b | 248 | enum nx_role role; /* Role. */ |
76ce9432 BP |
249 | struct hmap_node hmap_node; /* In struct ofproto's "controllers" map. */ |
250 | struct discovery *discovery; /* Controller discovery object, if enabled. */ | |
251 | struct status_category *ss; /* Switch status category. */ | |
d2ede7bc | 252 | enum ofproto_band band; /* In-band or out-of-band? */ |
064af421 BP |
253 | }; |
254 | ||
76ce9432 BP |
255 | /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's |
256 | * "schedulers" array. Their values are 0 and 1, and their meanings and values | |
257 | * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In | |
258 | * case anything ever changes, check their values here. */ | |
259 | #define N_SCHEDULERS 2 | |
260 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == 0); | |
261 | BUILD_ASSERT_DECL(OFPR_NO_MATCH == _ODPL_MISS_NR); | |
262 | BUILD_ASSERT_DECL(OFPR_ACTION == 1); | |
263 | BUILD_ASSERT_DECL(OFPR_ACTION == _ODPL_ACTION_NR); | |
264 | ||
265 | static struct ofconn *ofconn_create(struct ofproto *, struct rconn *, | |
266 | enum ofconn_type); | |
c475ae67 | 267 | static void ofconn_destroy(struct ofconn *); |
3269c562 | 268 | static void ofconn_run(struct ofconn *); |
064af421 | 269 | static void ofconn_wait(struct ofconn *); |
c91248b3 | 270 | static bool ofconn_receives_async_msgs(const struct ofconn *); |
eb15cdbb | 271 | static char *ofconn_make_name(const struct ofproto *, const char *target); |
7d674866 | 272 | static void ofconn_set_rate_limit(struct ofconn *, int rate, int burst); |
c91248b3 | 273 | |
064af421 BP |
274 | static void queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, |
275 | struct rconn_packet_counter *counter); | |
276 | ||
76ce9432 BP |
277 | static void send_packet_in(struct ofproto *, struct ofpbuf *odp_msg); |
278 | static void do_send_packet_in(struct ofpbuf *odp_msg, void *ofconn); | |
279 | ||
064af421 BP |
280 | struct ofproto { |
281 | /* Settings. */ | |
282 | uint64_t datapath_id; /* Datapath ID. */ | |
283 | uint64_t fallback_dpid; /* Datapath ID if no better choice found. */ | |
5a719c38 JP |
284 | char *mfr_desc; /* Manufacturer. */ |
285 | char *hw_desc; /* Hardware. */ | |
286 | char *sw_desc; /* Software version. */ | |
287 | char *serial_desc; /* Serial number. */ | |
8abc4ed7 | 288 | char *dp_desc; /* Datapath description. */ |
064af421 BP |
289 | |
290 | /* Datapath. */ | |
c228a364 | 291 | struct dpif *dpif; |
e9e28be3 | 292 | struct netdev_monitor *netdev_monitor; |
ca0f572c | 293 | struct hmap ports; /* Contains "struct ofport"s. */ |
064af421 BP |
294 | struct shash port_by_name; |
295 | uint32_t max_ports; | |
296 | ||
297 | /* Configuration. */ | |
298 | struct switch_status *switch_status; | |
064af421 | 299 | struct fail_open *fail_open; |
064af421 | 300 | struct netflow *netflow; |
72b06300 | 301 | struct ofproto_sflow *sflow; |
064af421 | 302 | |
d2ede7bc BP |
303 | /* In-band control. */ |
304 | struct in_band *in_band; | |
305 | long long int next_in_band_update; | |
917e50e1 BP |
306 | struct sockaddr_in *extra_in_band_remotes; |
307 | size_t n_extra_remotes; | |
b1da6250 | 308 | int in_band_queue; |
917e50e1 | 309 | |
064af421 BP |
310 | /* Flow table. */ |
311 | struct classifier cls; | |
064af421 | 312 | long long int next_expiration; |
bcf84111 BP |
313 | |
314 | /* Facets. */ | |
315 | struct hmap facets; | |
316 | bool need_revalidate; | |
064af421 BP |
317 | struct tag_set revalidate_set; |
318 | ||
319 | /* OpenFlow connections. */ | |
76ce9432 BP |
320 | struct hmap controllers; /* Controller "struct ofconn"s. */ |
321 | struct list all_conns; /* Contains "struct ofconn"s. */ | |
31681a5d | 322 | enum ofproto_fail_mode fail_mode; |
7d674866 BP |
323 | |
324 | /* OpenFlow listeners. */ | |
325 | struct hmap services; /* Contains "struct ofservice"s. */ | |
064af421 BP |
326 | struct pvconn **snoops; |
327 | size_t n_snoops; | |
328 | ||
329 | /* Hooks for ovs-vswitchd. */ | |
330 | const struct ofhooks *ofhooks; | |
331 | void *aux; | |
332 | ||
333 | /* Used by default ofhooks. */ | |
334 | struct mac_learning *ml; | |
335 | }; | |
336 | ||
337 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
338 | ||
339 | static const struct ofhooks default_ofhooks; | |
340 | ||
fa60c019 | 341 | static uint64_t pick_datapath_id(const struct ofproto *); |
064af421 | 342 | static uint64_t pick_fallback_dpid(void); |
76ce9432 | 343 | |
0de7a4b4 | 344 | static int ofproto_expire(struct ofproto *); |
4a4cdb3b | 345 | |
064af421 BP |
346 | static void handle_odp_msg(struct ofproto *, struct ofpbuf *); |
347 | ||
3269c562 | 348 | static void handle_openflow(struct ofconn *, struct ofpbuf *); |
064af421 | 349 | |
ca0f572c | 350 | static struct ofport *get_port(const struct ofproto *, uint16_t odp_port); |
064af421 BP |
351 | static void update_port(struct ofproto *, const char *devname); |
352 | static int init_ports(struct ofproto *); | |
353 | static void reinit_ports(struct ofproto *); | |
354 | ||
355 | int | |
1a6f1e2a JG |
356 | ofproto_create(const char *datapath, const char *datapath_type, |
357 | const struct ofhooks *ofhooks, void *aux, | |
064af421 BP |
358 | struct ofproto **ofprotop) |
359 | { | |
064af421 BP |
360 | struct odp_stats stats; |
361 | struct ofproto *p; | |
c228a364 | 362 | struct dpif *dpif; |
064af421 BP |
363 | int error; |
364 | ||
365 | *ofprotop = NULL; | |
366 | ||
367 | /* Connect to datapath and start listening for messages. */ | |
1a6f1e2a | 368 | error = dpif_open(datapath, datapath_type, &dpif); |
064af421 BP |
369 | if (error) { |
370 | VLOG_ERR("failed to open datapath %s: %s", datapath, strerror(error)); | |
371 | return error; | |
372 | } | |
c228a364 | 373 | error = dpif_get_dp_stats(dpif, &stats); |
064af421 BP |
374 | if (error) { |
375 | VLOG_ERR("failed to obtain stats for datapath %s: %s", | |
376 | datapath, strerror(error)); | |
c228a364 | 377 | dpif_close(dpif); |
064af421 BP |
378 | return error; |
379 | } | |
72b06300 | 380 | error = dpif_recv_set_mask(dpif, ODPL_MISS | ODPL_ACTION | ODPL_SFLOW); |
064af421 BP |
381 | if (error) { |
382 | VLOG_ERR("failed to listen on datapath %s: %s", | |
383 | datapath, strerror(error)); | |
c228a364 | 384 | dpif_close(dpif); |
064af421 BP |
385 | return error; |
386 | } | |
c228a364 | 387 | dpif_flow_flush(dpif); |
8f24562a | 388 | dpif_recv_purge(dpif); |
064af421 BP |
389 | |
390 | /* Initialize settings. */ | |
ec6fde61 | 391 | p = xzalloc(sizeof *p); |
064af421 | 392 | p->fallback_dpid = pick_fallback_dpid(); |
fa60c019 | 393 | p->datapath_id = p->fallback_dpid; |
5a719c38 JP |
394 | p->mfr_desc = xstrdup(DEFAULT_MFR_DESC); |
395 | p->hw_desc = xstrdup(DEFAULT_HW_DESC); | |
396 | p->sw_desc = xstrdup(DEFAULT_SW_DESC); | |
397 | p->serial_desc = xstrdup(DEFAULT_SERIAL_DESC); | |
23ff2821 | 398 | p->dp_desc = xstrdup(DEFAULT_DP_DESC); |
064af421 BP |
399 | |
400 | /* Initialize datapath. */ | |
401 | p->dpif = dpif; | |
8b61709d | 402 | p->netdev_monitor = netdev_monitor_create(); |
ca0f572c | 403 | hmap_init(&p->ports); |
064af421 BP |
404 | shash_init(&p->port_by_name); |
405 | p->max_ports = stats.max_ports; | |
406 | ||
407 | /* Initialize submodules. */ | |
408 | p->switch_status = switch_status_create(p); | |
064af421 | 409 | p->fail_open = NULL; |
064af421 | 410 | p->netflow = NULL; |
72b06300 | 411 | p->sflow = NULL; |
064af421 | 412 | |
b1da6250 BP |
413 | /* Initialize in-band control. */ |
414 | p->in_band = NULL; | |
415 | p->in_band_queue = -1; | |
416 | ||
064af421 BP |
417 | /* Initialize flow table. */ |
418 | classifier_init(&p->cls); | |
064af421 | 419 | p->next_expiration = time_msec() + 1000; |
bcf84111 BP |
420 | |
421 | /* Initialize facet table. */ | |
422 | hmap_init(&p->facets); | |
423 | p->need_revalidate = false; | |
064af421 BP |
424 | tag_set_init(&p->revalidate_set); |
425 | ||
426 | /* Initialize OpenFlow connections. */ | |
427 | list_init(&p->all_conns); | |
76ce9432 | 428 | hmap_init(&p->controllers); |
7d674866 | 429 | hmap_init(&p->services); |
064af421 BP |
430 | p->snoops = NULL; |
431 | p->n_snoops = 0; | |
432 | ||
433 | /* Initialize hooks. */ | |
434 | if (ofhooks) { | |
435 | p->ofhooks = ofhooks; | |
436 | p->aux = aux; | |
437 | p->ml = NULL; | |
438 | } else { | |
439 | p->ofhooks = &default_ofhooks; | |
440 | p->aux = p; | |
441 | p->ml = mac_learning_create(); | |
442 | } | |
443 | ||
fa60c019 BP |
444 | /* Pick final datapath ID. */ |
445 | p->datapath_id = pick_datapath_id(p); | |
b123cc3c | 446 | VLOG_INFO("using datapath ID %016"PRIx64, p->datapath_id); |
fa60c019 | 447 | |
064af421 BP |
448 | *ofprotop = p; |
449 | return 0; | |
450 | } | |
451 | ||
452 | void | |
453 | ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id) | |
454 | { | |
455 | uint64_t old_dpid = p->datapath_id; | |
fa60c019 | 456 | p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p); |
064af421 | 457 | if (p->datapath_id != old_dpid) { |
b123cc3c | 458 | VLOG_INFO("datapath ID changed to %016"PRIx64, p->datapath_id); |
76ce9432 BP |
459 | |
460 | /* Force all active connections to reconnect, since there is no way to | |
461 | * notify a controller that the datapath ID has changed. */ | |
fa05809b | 462 | ofproto_reconnect_controllers(p); |
064af421 BP |
463 | } |
464 | } | |
465 | ||
76ce9432 BP |
466 | static bool |
467 | is_discovery_controller(const struct ofproto_controller *c) | |
468 | { | |
469 | return !strcmp(c->target, "discover"); | |
470 | } | |
471 | ||
472 | static bool | |
473 | is_in_band_controller(const struct ofproto_controller *c) | |
474 | { | |
475 | return is_discovery_controller(c) || c->band == OFPROTO_IN_BAND; | |
476 | } | |
477 | ||
478 | /* Creates a new controller in 'ofproto'. Some of the settings are initially | |
479 | * drawn from 'c', but update_controller() needs to be called later to finish | |
480 | * the new ofconn's configuration. */ | |
481 | static void | |
482 | add_controller(struct ofproto *ofproto, const struct ofproto_controller *c) | |
483 | { | |
484 | struct discovery *discovery; | |
485 | struct ofconn *ofconn; | |
486 | ||
487 | if (is_discovery_controller(c)) { | |
488 | int error = discovery_create(c->accept_re, c->update_resolv_conf, | |
489 | ofproto->dpif, ofproto->switch_status, | |
490 | &discovery); | |
491 | if (error) { | |
492 | return; | |
493 | } | |
494 | } else { | |
495 | discovery = NULL; | |
496 | } | |
497 | ||
5899143f | 498 | ofconn = ofconn_create(ofproto, rconn_create(5, 8), OFCONN_PRIMARY); |
76ce9432 BP |
499 | ofconn->pktbuf = pktbuf_create(); |
500 | ofconn->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN; | |
501 | if (discovery) { | |
502 | ofconn->discovery = discovery; | |
503 | } else { | |
eb15cdbb BP |
504 | char *name = ofconn_make_name(ofproto, c->target); |
505 | rconn_connect(ofconn->rconn, c->target, name); | |
506 | free(name); | |
76ce9432 BP |
507 | } |
508 | hmap_insert(&ofproto->controllers, &ofconn->hmap_node, | |
509 | hash_string(c->target, 0)); | |
510 | } | |
511 | ||
512 | /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's | |
513 | * target or turn discovery on or off (these are done by creating new ofconns | |
514 | * and deleting old ones), but it can update the rest of an ofconn's | |
515 | * settings. */ | |
516 | static void | |
517 | update_controller(struct ofconn *ofconn, const struct ofproto_controller *c) | |
064af421 | 518 | { |
76ce9432 | 519 | int probe_interval; |
79c9f2ee | 520 | |
d2ede7bc BP |
521 | ofconn->band = (is_in_band_controller(c) |
522 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
523 | ||
76ce9432 | 524 | rconn_set_max_backoff(ofconn->rconn, c->max_backoff); |
79c9f2ee | 525 | |
76ce9432 BP |
526 | probe_interval = c->probe_interval ? MAX(c->probe_interval, 5) : 0; |
527 | rconn_set_probe_interval(ofconn->rconn, probe_interval); | |
79c9f2ee | 528 | |
76ce9432 BP |
529 | if (ofconn->discovery) { |
530 | discovery_set_update_resolv_conf(ofconn->discovery, | |
531 | c->update_resolv_conf); | |
532 | discovery_set_accept_controller_re(ofconn->discovery, c->accept_re); | |
533 | } | |
79c9f2ee | 534 | |
7d674866 | 535 | ofconn_set_rate_limit(ofconn, c->rate_limit, c->burst_limit); |
76ce9432 | 536 | } |
79c9f2ee | 537 | |
76ce9432 BP |
538 | static const char * |
539 | ofconn_get_target(const struct ofconn *ofconn) | |
540 | { | |
eb15cdbb | 541 | return ofconn->discovery ? "discover" : rconn_get_target(ofconn->rconn); |
76ce9432 BP |
542 | } |
543 | ||
544 | static struct ofconn * | |
545 | find_controller_by_target(struct ofproto *ofproto, const char *target) | |
546 | { | |
547 | struct ofconn *ofconn; | |
548 | ||
4e8e4213 | 549 | HMAP_FOR_EACH_WITH_HASH (ofconn, hmap_node, |
76ce9432 BP |
550 | hash_string(target, 0), &ofproto->controllers) { |
551 | if (!strcmp(ofconn_get_target(ofconn), target)) { | |
552 | return ofconn; | |
79c9f2ee | 553 | } |
064af421 | 554 | } |
76ce9432 BP |
555 | return NULL; |
556 | } | |
064af421 | 557 | |
d2ede7bc BP |
558 | static void |
559 | update_in_band_remotes(struct ofproto *ofproto) | |
560 | { | |
561 | const struct ofconn *ofconn; | |
562 | struct sockaddr_in *addrs; | |
917e50e1 | 563 | size_t max_addrs, n_addrs; |
d2ede7bc | 564 | bool discovery; |
917e50e1 | 565 | size_t i; |
d2ede7bc | 566 | |
917e50e1 BP |
567 | /* Allocate enough memory for as many remotes as we could possibly have. */ |
568 | max_addrs = ofproto->n_extra_remotes + hmap_count(&ofproto->controllers); | |
569 | addrs = xmalloc(max_addrs * sizeof *addrs); | |
d2ede7bc BP |
570 | n_addrs = 0; |
571 | ||
572 | /* Add all the remotes. */ | |
573 | discovery = false; | |
4e8e4213 | 574 | HMAP_FOR_EACH (ofconn, hmap_node, &ofproto->controllers) { |
d2ede7bc BP |
575 | struct sockaddr_in *sin = &addrs[n_addrs]; |
576 | ||
487ec65f BP |
577 | if (ofconn->band == OFPROTO_OUT_OF_BAND) { |
578 | continue; | |
579 | } | |
580 | ||
d2ede7bc BP |
581 | sin->sin_addr.s_addr = rconn_get_remote_ip(ofconn->rconn); |
582 | if (sin->sin_addr.s_addr) { | |
583 | sin->sin_port = rconn_get_remote_port(ofconn->rconn); | |
584 | n_addrs++; | |
585 | } | |
586 | if (ofconn->discovery) { | |
587 | discovery = true; | |
588 | } | |
589 | } | |
917e50e1 BP |
590 | for (i = 0; i < ofproto->n_extra_remotes; i++) { |
591 | addrs[n_addrs++] = ofproto->extra_in_band_remotes[i]; | |
592 | } | |
d2ede7bc BP |
593 | |
594 | /* Create or update or destroy in-band. | |
595 | * | |
596 | * Ordinarily we only enable in-band if there's at least one remote | |
597 | * address, but discovery needs the in-band rules for DHCP to be installed | |
598 | * even before we know any remote addresses. */ | |
599 | if (n_addrs || discovery) { | |
600 | if (!ofproto->in_band) { | |
601 | in_band_create(ofproto, ofproto->dpif, ofproto->switch_status, | |
602 | &ofproto->in_band); | |
603 | } | |
40cae670 BP |
604 | if (ofproto->in_band) { |
605 | in_band_set_remotes(ofproto->in_band, addrs, n_addrs); | |
606 | } | |
b1da6250 | 607 | in_band_set_queue(ofproto->in_band, ofproto->in_band_queue); |
d2ede7bc BP |
608 | ofproto->next_in_band_update = time_msec() + 1000; |
609 | } else { | |
610 | in_band_destroy(ofproto->in_band); | |
611 | ofproto->in_band = NULL; | |
612 | } | |
613 | ||
614 | /* Clean up. */ | |
615 | free(addrs); | |
616 | } | |
617 | ||
31681a5d JP |
618 | static void |
619 | update_fail_open(struct ofproto *p) | |
620 | { | |
621 | struct ofconn *ofconn; | |
622 | ||
623 | if (!hmap_is_empty(&p->controllers) | |
624 | && p->fail_mode == OFPROTO_FAIL_STANDALONE) { | |
625 | struct rconn **rconns; | |
626 | size_t n; | |
627 | ||
628 | if (!p->fail_open) { | |
629 | p->fail_open = fail_open_create(p, p->switch_status); | |
630 | } | |
631 | ||
632 | n = 0; | |
633 | rconns = xmalloc(hmap_count(&p->controllers) * sizeof *rconns); | |
4e8e4213 | 634 | HMAP_FOR_EACH (ofconn, hmap_node, &p->controllers) { |
31681a5d JP |
635 | rconns[n++] = ofconn->rconn; |
636 | } | |
637 | ||
638 | fail_open_set_controllers(p->fail_open, rconns, n); | |
639 | /* p->fail_open takes ownership of 'rconns'. */ | |
640 | } else { | |
641 | fail_open_destroy(p->fail_open); | |
642 | p->fail_open = NULL; | |
643 | } | |
644 | } | |
645 | ||
76ce9432 BP |
646 | void |
647 | ofproto_set_controllers(struct ofproto *p, | |
648 | const struct ofproto_controller *controllers, | |
649 | size_t n_controllers) | |
650 | { | |
651 | struct shash new_controllers; | |
7d674866 BP |
652 | struct ofconn *ofconn, *next_ofconn; |
653 | struct ofservice *ofservice, *next_ofservice; | |
76ce9432 | 654 | bool ss_exists; |
76ce9432 | 655 | size_t i; |
79c9f2ee | 656 | |
7d674866 BP |
657 | /* Create newly configured controllers and services. |
658 | * Create a name to ofproto_controller mapping in 'new_controllers'. */ | |
76ce9432 BP |
659 | shash_init(&new_controllers); |
660 | for (i = 0; i < n_controllers; i++) { | |
661 | const struct ofproto_controller *c = &controllers[i]; | |
662 | ||
7d674866 BP |
663 | if (!vconn_verify_name(c->target) || !strcmp(c->target, "discover")) { |
664 | if (!find_controller_by_target(p, c->target)) { | |
665 | add_controller(p, c); | |
666 | } | |
667 | } else if (!pvconn_verify_name(c->target)) { | |
668 | if (!ofservice_lookup(p, c->target) && ofservice_create(p, c)) { | |
669 | continue; | |
670 | } | |
671 | } else { | |
672 | VLOG_WARN_RL(&rl, "%s: unsupported controller \"%s\"", | |
673 | dpif_name(p->dpif), c->target); | |
674 | continue; | |
76ce9432 | 675 | } |
7d674866 BP |
676 | |
677 | shash_add_once(&new_controllers, c->target, &controllers[i]); | |
76ce9432 BP |
678 | } |
679 | ||
7d674866 BP |
680 | /* Delete controllers that are no longer configured. |
681 | * Update configuration of all now-existing controllers. */ | |
76ce9432 | 682 | ss_exists = false; |
4e8e4213 | 683 | HMAP_FOR_EACH_SAFE (ofconn, next_ofconn, hmap_node, &p->controllers) { |
76ce9432 BP |
684 | struct ofproto_controller *c; |
685 | ||
686 | c = shash_find_data(&new_controllers, ofconn_get_target(ofconn)); | |
687 | if (!c) { | |
688 | ofconn_destroy(ofconn); | |
79c9f2ee | 689 | } else { |
76ce9432 | 690 | update_controller(ofconn, c); |
76ce9432 BP |
691 | if (ofconn->ss) { |
692 | ss_exists = true; | |
693 | } | |
76ce9432 BP |
694 | } |
695 | } | |
7d674866 BP |
696 | |
697 | /* Delete services that are no longer configured. | |
698 | * Update configuration of all now-existing services. */ | |
4e8e4213 | 699 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 BP |
700 | struct ofproto_controller *c; |
701 | ||
702 | c = shash_find_data(&new_controllers, | |
703 | pvconn_get_name(ofservice->pvconn)); | |
704 | if (!c) { | |
705 | ofservice_destroy(p, ofservice); | |
706 | } else { | |
707 | ofservice_reconfigure(ofservice, c); | |
708 | } | |
709 | } | |
710 | ||
76ce9432 BP |
711 | shash_destroy(&new_controllers); |
712 | ||
d2ede7bc | 713 | update_in_band_remotes(p); |
31681a5d | 714 | update_fail_open(p); |
79c9f2ee | 715 | |
76ce9432 BP |
716 | if (!hmap_is_empty(&p->controllers) && !ss_exists) { |
717 | ofconn = CONTAINER_OF(hmap_first(&p->controllers), | |
718 | struct ofconn, hmap_node); | |
719 | ofconn->ss = switch_status_register(p->switch_status, "remote", | |
720 | rconn_status_cb, ofconn->rconn); | |
79c9f2ee | 721 | } |
064af421 BP |
722 | } |
723 | ||
31681a5d JP |
724 | void |
725 | ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode) | |
726 | { | |
727 | p->fail_mode = fail_mode; | |
728 | update_fail_open(p); | |
729 | } | |
730 | ||
fa05809b BP |
731 | /* Drops the connections between 'ofproto' and all of its controllers, forcing |
732 | * them to reconnect. */ | |
733 | void | |
734 | ofproto_reconnect_controllers(struct ofproto *ofproto) | |
735 | { | |
736 | struct ofconn *ofconn; | |
737 | ||
4e8e4213 | 738 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
fa05809b BP |
739 | rconn_reconnect(ofconn->rconn); |
740 | } | |
741 | } | |
742 | ||
917e50e1 BP |
743 | static bool |
744 | any_extras_changed(const struct ofproto *ofproto, | |
745 | const struct sockaddr_in *extras, size_t n) | |
746 | { | |
747 | size_t i; | |
748 | ||
749 | if (n != ofproto->n_extra_remotes) { | |
750 | return true; | |
751 | } | |
752 | ||
753 | for (i = 0; i < n; i++) { | |
754 | const struct sockaddr_in *old = &ofproto->extra_in_band_remotes[i]; | |
755 | const struct sockaddr_in *new = &extras[i]; | |
756 | ||
757 | if (old->sin_addr.s_addr != new->sin_addr.s_addr || | |
758 | old->sin_port != new->sin_port) { | |
759 | return true; | |
760 | } | |
761 | } | |
762 | ||
763 | return false; | |
764 | } | |
765 | ||
766 | /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s | |
767 | * in-band control should guarantee access, in the same way that in-band | |
768 | * control guarantees access to OpenFlow controllers. */ | |
769 | void | |
770 | ofproto_set_extra_in_band_remotes(struct ofproto *ofproto, | |
771 | const struct sockaddr_in *extras, size_t n) | |
772 | { | |
773 | if (!any_extras_changed(ofproto, extras, n)) { | |
774 | return; | |
775 | } | |
776 | ||
777 | free(ofproto->extra_in_band_remotes); | |
778 | ofproto->n_extra_remotes = n; | |
779 | ofproto->extra_in_band_remotes = xmemdup(extras, n * sizeof *extras); | |
780 | ||
781 | update_in_band_remotes(ofproto); | |
782 | } | |
783 | ||
b1da6250 BP |
784 | /* Sets the OpenFlow queue used by flows set up by in-band control on |
785 | * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control | |
786 | * flows will use the default queue. */ | |
787 | void | |
788 | ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id) | |
789 | { | |
790 | if (queue_id != ofproto->in_band_queue) { | |
791 | ofproto->in_band_queue = queue_id; | |
792 | update_in_band_remotes(ofproto); | |
793 | } | |
794 | } | |
795 | ||
064af421 BP |
796 | void |
797 | ofproto_set_desc(struct ofproto *p, | |
5a719c38 JP |
798 | const char *mfr_desc, const char *hw_desc, |
799 | const char *sw_desc, const char *serial_desc, | |
8abc4ed7 | 800 | const char *dp_desc) |
064af421 | 801 | { |
5a719c38 JP |
802 | struct ofp_desc_stats *ods; |
803 | ||
804 | if (mfr_desc) { | |
805 | if (strlen(mfr_desc) >= sizeof ods->mfr_desc) { | |
806 | VLOG_WARN("truncating mfr_desc, must be less than %zu characters", | |
807 | sizeof ods->mfr_desc); | |
808 | } | |
809 | free(p->mfr_desc); | |
810 | p->mfr_desc = xstrdup(mfr_desc); | |
064af421 | 811 | } |
5a719c38 JP |
812 | if (hw_desc) { |
813 | if (strlen(hw_desc) >= sizeof ods->hw_desc) { | |
814 | VLOG_WARN("truncating hw_desc, must be less than %zu characters", | |
815 | sizeof ods->hw_desc); | |
816 | } | |
817 | free(p->hw_desc); | |
818 | p->hw_desc = xstrdup(hw_desc); | |
064af421 | 819 | } |
5a719c38 JP |
820 | if (sw_desc) { |
821 | if (strlen(sw_desc) >= sizeof ods->sw_desc) { | |
822 | VLOG_WARN("truncating sw_desc, must be less than %zu characters", | |
823 | sizeof ods->sw_desc); | |
824 | } | |
825 | free(p->sw_desc); | |
826 | p->sw_desc = xstrdup(sw_desc); | |
827 | } | |
828 | if (serial_desc) { | |
829 | if (strlen(serial_desc) >= sizeof ods->serial_num) { | |
830 | VLOG_WARN("truncating serial_desc, must be less than %zu " | |
831 | "characters", | |
832 | sizeof ods->serial_num); | |
833 | } | |
834 | free(p->serial_desc); | |
835 | p->serial_desc = xstrdup(serial_desc); | |
064af421 | 836 | } |
8abc4ed7 | 837 | if (dp_desc) { |
5a719c38 JP |
838 | if (strlen(dp_desc) >= sizeof ods->dp_desc) { |
839 | VLOG_WARN("truncating dp_desc, must be less than %zu characters", | |
840 | sizeof ods->dp_desc); | |
841 | } | |
8abc4ed7 JP |
842 | free(p->dp_desc); |
843 | p->dp_desc = xstrdup(dp_desc); | |
844 | } | |
064af421 BP |
845 | } |
846 | ||
064af421 BP |
847 | static int |
848 | set_pvconns(struct pvconn ***pvconnsp, size_t *n_pvconnsp, | |
849 | const struct svec *svec) | |
850 | { | |
851 | struct pvconn **pvconns = *pvconnsp; | |
852 | size_t n_pvconns = *n_pvconnsp; | |
853 | int retval = 0; | |
854 | size_t i; | |
855 | ||
856 | for (i = 0; i < n_pvconns; i++) { | |
857 | pvconn_close(pvconns[i]); | |
858 | } | |
859 | free(pvconns); | |
860 | ||
861 | pvconns = xmalloc(svec->n * sizeof *pvconns); | |
862 | n_pvconns = 0; | |
863 | for (i = 0; i < svec->n; i++) { | |
864 | const char *name = svec->names[i]; | |
865 | struct pvconn *pvconn; | |
866 | int error; | |
867 | ||
868 | error = pvconn_open(name, &pvconn); | |
869 | if (!error) { | |
870 | pvconns[n_pvconns++] = pvconn; | |
871 | } else { | |
872 | VLOG_ERR("failed to listen on %s: %s", name, strerror(error)); | |
873 | if (!retval) { | |
874 | retval = error; | |
875 | } | |
876 | } | |
877 | } | |
878 | ||
879 | *pvconnsp = pvconns; | |
880 | *n_pvconnsp = n_pvconns; | |
881 | ||
882 | return retval; | |
883 | } | |
884 | ||
064af421 BP |
885 | int |
886 | ofproto_set_snoops(struct ofproto *ofproto, const struct svec *snoops) | |
887 | { | |
888 | return set_pvconns(&ofproto->snoops, &ofproto->n_snoops, snoops); | |
889 | } | |
890 | ||
891 | int | |
0193b2af JG |
892 | ofproto_set_netflow(struct ofproto *ofproto, |
893 | const struct netflow_options *nf_options) | |
064af421 | 894 | { |
76343538 | 895 | if (nf_options && nf_options->collectors.n) { |
064af421 BP |
896 | if (!ofproto->netflow) { |
897 | ofproto->netflow = netflow_create(); | |
898 | } | |
0193b2af | 899 | return netflow_set_options(ofproto->netflow, nf_options); |
064af421 BP |
900 | } else { |
901 | netflow_destroy(ofproto->netflow); | |
902 | ofproto->netflow = NULL; | |
903 | return 0; | |
904 | } | |
905 | } | |
906 | ||
72b06300 BP |
907 | void |
908 | ofproto_set_sflow(struct ofproto *ofproto, | |
909 | const struct ofproto_sflow_options *oso) | |
910 | { | |
911 | struct ofproto_sflow *os = ofproto->sflow; | |
912 | if (oso) { | |
913 | if (!os) { | |
914 | struct ofport *ofport; | |
72b06300 BP |
915 | |
916 | os = ofproto->sflow = ofproto_sflow_create(ofproto->dpif); | |
4e8e4213 | 917 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { |
ca0f572c | 918 | ofproto_sflow_add_port(os, ofport->odp_port, |
72b06300 BP |
919 | netdev_get_name(ofport->netdev)); |
920 | } | |
921 | } | |
922 | ofproto_sflow_set_options(os, oso); | |
923 | } else { | |
924 | ofproto_sflow_destroy(os); | |
925 | ofproto->sflow = NULL; | |
926 | } | |
927 | } | |
928 | ||
064af421 BP |
929 | uint64_t |
930 | ofproto_get_datapath_id(const struct ofproto *ofproto) | |
931 | { | |
932 | return ofproto->datapath_id; | |
933 | } | |
934 | ||
76ce9432 | 935 | bool |
7d674866 | 936 | ofproto_has_primary_controller(const struct ofproto *ofproto) |
064af421 | 937 | { |
76ce9432 | 938 | return !hmap_is_empty(&ofproto->controllers); |
064af421 BP |
939 | } |
940 | ||
abdfe474 JP |
941 | enum ofproto_fail_mode |
942 | ofproto_get_fail_mode(const struct ofproto *p) | |
943 | { | |
944 | return p->fail_mode; | |
945 | } | |
946 | ||
064af421 BP |
947 | void |
948 | ofproto_get_snoops(const struct ofproto *ofproto, struct svec *snoops) | |
949 | { | |
950 | size_t i; | |
951 | ||
952 | for (i = 0; i < ofproto->n_snoops; i++) { | |
953 | svec_add(snoops, pvconn_get_name(ofproto->snoops[i])); | |
954 | } | |
955 | } | |
956 | ||
957 | void | |
958 | ofproto_destroy(struct ofproto *p) | |
959 | { | |
7d674866 | 960 | struct ofservice *ofservice, *next_ofservice; |
064af421 | 961 | struct ofconn *ofconn, *next_ofconn; |
ca0f572c | 962 | struct ofport *ofport, *next_ofport; |
064af421 BP |
963 | size_t i; |
964 | ||
965 | if (!p) { | |
966 | return; | |
967 | } | |
968 | ||
f7de2cdf | 969 | /* Destroy fail-open and in-band early, since they touch the classifier. */ |
79c9f2ee BP |
970 | fail_open_destroy(p->fail_open); |
971 | p->fail_open = NULL; | |
972 | ||
973 | in_band_destroy(p->in_band); | |
974 | p->in_band = NULL; | |
917e50e1 | 975 | free(p->extra_in_band_remotes); |
2f6d3445 | 976 | |
064af421 BP |
977 | ofproto_flush_flows(p); |
978 | classifier_destroy(&p->cls); | |
bcf84111 | 979 | hmap_destroy(&p->facets); |
064af421 | 980 | |
4e8e4213 | 981 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
c475ae67 | 982 | ofconn_destroy(ofconn); |
064af421 | 983 | } |
76ce9432 | 984 | hmap_destroy(&p->controllers); |
064af421 | 985 | |
c228a364 | 986 | dpif_close(p->dpif); |
e9e28be3 | 987 | netdev_monitor_destroy(p->netdev_monitor); |
4e8e4213 | 988 | HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) { |
ca0f572c | 989 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
990 | ofport_free(ofport); |
991 | } | |
992 | shash_destroy(&p->port_by_name); | |
993 | ||
994 | switch_status_destroy(p->switch_status); | |
064af421 | 995 | netflow_destroy(p->netflow); |
72b06300 | 996 | ofproto_sflow_destroy(p->sflow); |
064af421 | 997 | |
4e8e4213 | 998 | HMAP_FOR_EACH_SAFE (ofservice, next_ofservice, node, &p->services) { |
7d674866 | 999 | ofservice_destroy(p, ofservice); |
064af421 | 1000 | } |
7d674866 | 1001 | hmap_destroy(&p->services); |
064af421 BP |
1002 | |
1003 | for (i = 0; i < p->n_snoops; i++) { | |
1004 | pvconn_close(p->snoops[i]); | |
1005 | } | |
1006 | free(p->snoops); | |
1007 | ||
1008 | mac_learning_destroy(p->ml); | |
1009 | ||
5a719c38 JP |
1010 | free(p->mfr_desc); |
1011 | free(p->hw_desc); | |
1012 | free(p->sw_desc); | |
1013 | free(p->serial_desc); | |
cb871ae0 JP |
1014 | free(p->dp_desc); |
1015 | ||
ca0f572c | 1016 | hmap_destroy(&p->ports); |
3b917492 | 1017 | |
064af421 BP |
1018 | free(p); |
1019 | } | |
1020 | ||
1021 | int | |
1022 | ofproto_run(struct ofproto *p) | |
1023 | { | |
1024 | int error = ofproto_run1(p); | |
1025 | if (!error) { | |
1026 | error = ofproto_run2(p, false); | |
1027 | } | |
1028 | return error; | |
1029 | } | |
1030 | ||
e9e28be3 BP |
1031 | static void |
1032 | process_port_change(struct ofproto *ofproto, int error, char *devname) | |
1033 | { | |
1034 | if (error == ENOBUFS) { | |
1035 | reinit_ports(ofproto); | |
1036 | } else if (!error) { | |
1037 | update_port(ofproto, devname); | |
1038 | free(devname); | |
1039 | } | |
1040 | } | |
1041 | ||
e2bfacb6 BP |
1042 | /* Returns a "preference level" for snooping 'ofconn'. A higher return value |
1043 | * means that 'ofconn' is more interesting for monitoring than a lower return | |
1044 | * value. */ | |
1045 | static int | |
1046 | snoop_preference(const struct ofconn *ofconn) | |
1047 | { | |
1048 | switch (ofconn->role) { | |
1049 | case NX_ROLE_MASTER: | |
1050 | return 3; | |
1051 | case NX_ROLE_OTHER: | |
1052 | return 2; | |
1053 | case NX_ROLE_SLAVE: | |
1054 | return 1; | |
1055 | default: | |
1056 | /* Shouldn't happen. */ | |
1057 | return 0; | |
1058 | } | |
1059 | } | |
1060 | ||
76ce9432 BP |
1061 | /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'. |
1062 | * Connects this vconn to a controller. */ | |
1063 | static void | |
1064 | add_snooper(struct ofproto *ofproto, struct vconn *vconn) | |
1065 | { | |
e2bfacb6 | 1066 | struct ofconn *ofconn, *best; |
76ce9432 | 1067 | |
e2bfacb6 BP |
1068 | /* Pick a controller for monitoring. */ |
1069 | best = NULL; | |
4e8e4213 | 1070 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
5899143f | 1071 | if (ofconn->type == OFCONN_PRIMARY |
e2bfacb6 BP |
1072 | && (!best || snoop_preference(ofconn) > snoop_preference(best))) { |
1073 | best = ofconn; | |
76ce9432 | 1074 | } |
e2bfacb6 | 1075 | } |
76ce9432 | 1076 | |
e2bfacb6 BP |
1077 | if (best) { |
1078 | rconn_add_monitor(best->rconn, vconn); | |
1079 | } else { | |
1080 | VLOG_INFO_RL(&rl, "no controller connection to snoop"); | |
1081 | vconn_close(vconn); | |
76ce9432 | 1082 | } |
76ce9432 BP |
1083 | } |
1084 | ||
064af421 BP |
1085 | int |
1086 | ofproto_run1(struct ofproto *p) | |
1087 | { | |
1088 | struct ofconn *ofconn, *next_ofconn; | |
7d674866 | 1089 | struct ofservice *ofservice; |
064af421 BP |
1090 | char *devname; |
1091 | int error; | |
1092 | int i; | |
1093 | ||
149f577a JG |
1094 | if (shash_is_empty(&p->port_by_name)) { |
1095 | init_ports(p); | |
1096 | } | |
1097 | ||
064af421 BP |
1098 | for (i = 0; i < 50; i++) { |
1099 | struct ofpbuf *buf; | |
064af421 | 1100 | |
c228a364 | 1101 | error = dpif_recv(p->dpif, &buf); |
064af421 BP |
1102 | if (error) { |
1103 | if (error == ENODEV) { | |
1104 | /* Someone destroyed the datapath behind our back. The caller | |
1105 | * better destroy us and give up, because we're just going to | |
1106 | * spin from here on out. */ | |
39a559f2 BP |
1107 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 5); |
1108 | VLOG_ERR_RL(&rl2, "%s: datapath was destroyed externally", | |
c228a364 | 1109 | dpif_name(p->dpif)); |
064af421 BP |
1110 | return ENODEV; |
1111 | } | |
1112 | break; | |
1113 | } | |
1114 | ||
1115 | handle_odp_msg(p, buf); | |
1116 | } | |
1117 | ||
e9e28be3 BP |
1118 | while ((error = dpif_port_poll(p->dpif, &devname)) != EAGAIN) { |
1119 | process_port_change(p, error, devname); | |
1120 | } | |
1121 | while ((error = netdev_monitor_poll(p->netdev_monitor, | |
1122 | &devname)) != EAGAIN) { | |
1123 | process_port_change(p, error, devname); | |
064af421 BP |
1124 | } |
1125 | ||
1126 | if (p->in_band) { | |
d2ede7bc BP |
1127 | if (time_msec() >= p->next_in_band_update) { |
1128 | update_in_band_remotes(p); | |
1129 | } | |
064af421 BP |
1130 | in_band_run(p->in_band); |
1131 | } | |
064af421 | 1132 | |
4e8e4213 | 1133 | LIST_FOR_EACH_SAFE (ofconn, next_ofconn, node, &p->all_conns) { |
3269c562 | 1134 | ofconn_run(ofconn); |
064af421 BP |
1135 | } |
1136 | ||
7778bd15 BP |
1137 | /* Fail-open maintenance. Do this after processing the ofconns since |
1138 | * fail-open checks the status of the controller rconn. */ | |
1139 | if (p->fail_open) { | |
1140 | fail_open_run(p->fail_open); | |
1141 | } | |
1142 | ||
4e8e4213 | 1143 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
064af421 BP |
1144 | struct vconn *vconn; |
1145 | int retval; | |
1146 | ||
7d674866 | 1147 | retval = pvconn_accept(ofservice->pvconn, OFP_VERSION, &vconn); |
064af421 | 1148 | if (!retval) { |
9794e806 | 1149 | struct rconn *rconn; |
eb15cdbb | 1150 | char *name; |
9794e806 | 1151 | |
7d674866 | 1152 | rconn = rconn_create(ofservice->probe_interval, 0); |
eb15cdbb BP |
1153 | name = ofconn_make_name(p, vconn_get_name(vconn)); |
1154 | rconn_connect_unreliably(rconn, vconn, name); | |
1155 | free(name); | |
1156 | ||
7d674866 BP |
1157 | ofconn = ofconn_create(p, rconn, OFCONN_SERVICE); |
1158 | ofconn_set_rate_limit(ofconn, ofservice->rate_limit, | |
1159 | ofservice->burst_limit); | |
064af421 BP |
1160 | } else if (retval != EAGAIN) { |
1161 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | for (i = 0; i < p->n_snoops; i++) { | |
1166 | struct vconn *vconn; | |
1167 | int retval; | |
1168 | ||
1169 | retval = pvconn_accept(p->snoops[i], OFP_VERSION, &vconn); | |
1170 | if (!retval) { | |
76ce9432 | 1171 | add_snooper(p, vconn); |
064af421 BP |
1172 | } else if (retval != EAGAIN) { |
1173 | VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval)); | |
1174 | } | |
1175 | } | |
1176 | ||
1177 | if (time_msec() >= p->next_expiration) { | |
0de7a4b4 BP |
1178 | int delay = ofproto_expire(p); |
1179 | p->next_expiration = time_msec() + delay; | |
064af421 | 1180 | COVERAGE_INC(ofproto_expiration); |
064af421 BP |
1181 | } |
1182 | ||
1183 | if (p->netflow) { | |
1184 | netflow_run(p->netflow); | |
1185 | } | |
72b06300 BP |
1186 | if (p->sflow) { |
1187 | ofproto_sflow_run(p->sflow); | |
1188 | } | |
064af421 BP |
1189 | |
1190 | return 0; | |
1191 | } | |
1192 | ||
064af421 BP |
1193 | int |
1194 | ofproto_run2(struct ofproto *p, bool revalidate_all) | |
1195 | { | |
bcf84111 BP |
1196 | /* Figure out what we need to revalidate now, if anything. */ |
1197 | struct tag_set revalidate_set = p->revalidate_set; | |
1198 | if (p->need_revalidate) { | |
1199 | revalidate_all = true; | |
1200 | } | |
1201 | ||
1202 | /* Clear the revalidation flags. */ | |
1203 | tag_set_init(&p->revalidate_set); | |
1204 | p->need_revalidate = false; | |
1205 | ||
1206 | /* Now revalidate if there's anything to do. */ | |
1207 | if (revalidate_all || !tag_set_is_empty(&revalidate_set)) { | |
1208 | struct facet *facet, *next; | |
1209 | ||
1210 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &p->facets) { | |
1211 | if (revalidate_all | |
1212 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
1213 | facet_revalidate(p, facet); | |
1214 | } | |
1215 | } | |
064af421 BP |
1216 | } |
1217 | ||
1218 | return 0; | |
1219 | } | |
1220 | ||
1221 | void | |
1222 | ofproto_wait(struct ofproto *p) | |
1223 | { | |
7d674866 | 1224 | struct ofservice *ofservice; |
064af421 BP |
1225 | struct ofconn *ofconn; |
1226 | size_t i; | |
1227 | ||
c228a364 | 1228 | dpif_recv_wait(p->dpif); |
e9e28be3 BP |
1229 | dpif_port_poll_wait(p->dpif); |
1230 | netdev_monitor_poll_wait(p->netdev_monitor); | |
4e8e4213 | 1231 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1232 | ofconn_wait(ofconn); |
1233 | } | |
1234 | if (p->in_band) { | |
7cf8b266 | 1235 | poll_timer_wait_until(p->next_in_band_update); |
064af421 BP |
1236 | in_band_wait(p->in_band); |
1237 | } | |
064af421 BP |
1238 | if (p->fail_open) { |
1239 | fail_open_wait(p->fail_open); | |
1240 | } | |
72b06300 BP |
1241 | if (p->sflow) { |
1242 | ofproto_sflow_wait(p->sflow); | |
1243 | } | |
064af421 BP |
1244 | if (!tag_set_is_empty(&p->revalidate_set)) { |
1245 | poll_immediate_wake(); | |
1246 | } | |
1247 | if (p->need_revalidate) { | |
1248 | /* Shouldn't happen, but if it does just go around again. */ | |
1249 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
1250 | poll_immediate_wake(); | |
1251 | } else if (p->next_expiration != LLONG_MAX) { | |
7cf8b266 | 1252 | poll_timer_wait_until(p->next_expiration); |
064af421 | 1253 | } |
4e8e4213 | 1254 | HMAP_FOR_EACH (ofservice, node, &p->services) { |
7d674866 | 1255 | pvconn_wait(ofservice->pvconn); |
064af421 BP |
1256 | } |
1257 | for (i = 0; i < p->n_snoops; i++) { | |
1258 | pvconn_wait(p->snoops[i]); | |
1259 | } | |
1260 | } | |
1261 | ||
1262 | void | |
1263 | ofproto_revalidate(struct ofproto *ofproto, tag_type tag) | |
1264 | { | |
1265 | tag_set_add(&ofproto->revalidate_set, tag); | |
1266 | } | |
1267 | ||
1268 | struct tag_set * | |
1269 | ofproto_get_revalidate_set(struct ofproto *ofproto) | |
1270 | { | |
1271 | return &ofproto->revalidate_set; | |
1272 | } | |
1273 | ||
1274 | bool | |
1275 | ofproto_is_alive(const struct ofproto *p) | |
1276 | { | |
76ce9432 | 1277 | return !hmap_is_empty(&p->controllers); |
064af421 BP |
1278 | } |
1279 | ||
3a6ccc8c BP |
1280 | /* Deletes port number 'odp_port' from the datapath for 'ofproto'. |
1281 | * | |
1282 | * This is almost the same as calling dpif_port_del() directly on the | |
1283 | * datapath, but it also makes 'ofproto' close its open netdev for the port | |
1284 | * (if any). This makes it possible to create a new netdev of a different | |
1285 | * type under the same name, which otherwise the netdev library would refuse | |
1286 | * to do because of the conflict. (The netdev would eventually get closed on | |
1287 | * the next trip through ofproto_run(), but this interface is more direct.) | |
1288 | * | |
3a6ccc8c BP |
1289 | * Returns 0 if successful, otherwise a positive errno. */ |
1290 | int | |
1291 | ofproto_port_del(struct ofproto *ofproto, uint16_t odp_port) | |
1292 | { | |
1293 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1294 | const char *name = ofport ? (char *) ofport->opp.name : "<unknown>"; | |
1295 | int error; | |
1296 | ||
1297 | error = dpif_port_del(ofproto->dpif, odp_port); | |
1298 | if (error) { | |
1299 | VLOG_ERR("%s: failed to remove port %"PRIu16" (%s) interface (%s)", | |
1300 | dpif_name(ofproto->dpif), odp_port, name, strerror(error)); | |
1301 | } else if (ofport) { | |
1302 | /* 'name' is ofport->opp.name and update_port() is going to destroy | |
1303 | * 'ofport'. Just in case update_port() refers to 'name' after it | |
1304 | * destroys 'ofport', make a copy of it around the update_port() | |
1305 | * call. */ | |
1306 | char *devname = xstrdup(name); | |
1307 | update_port(ofproto, devname); | |
1308 | free(devname); | |
1309 | } | |
1310 | return error; | |
1311 | } | |
1312 | ||
a4e2e1f2 EJ |
1313 | /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns |
1314 | * true if 'odp_port' exists and should be included, false otherwise. */ | |
1315 | bool | |
1316 | ofproto_port_is_floodable(struct ofproto *ofproto, uint16_t odp_port) | |
1317 | { | |
1318 | struct ofport *ofport = get_port(ofproto, odp_port); | |
1319 | return ofport && !(ofport->opp.config & OFPPC_NO_FLOOD); | |
1320 | } | |
1321 | ||
064af421 | 1322 | int |
ae412e7d | 1323 | ofproto_send_packet(struct ofproto *p, const struct flow *flow, |
064af421 BP |
1324 | const union ofp_action *actions, size_t n_actions, |
1325 | const struct ofpbuf *packet) | |
1326 | { | |
1327 | struct odp_actions odp_actions; | |
1328 | int error; | |
1329 | ||
1330 | error = xlate_actions(actions, n_actions, flow, p, packet, &odp_actions, | |
6a07af36 | 1331 | NULL, NULL, NULL); |
064af421 BP |
1332 | if (error) { |
1333 | return error; | |
1334 | } | |
1335 | ||
1336 | /* XXX Should we translate the dpif_execute() errno value into an OpenFlow | |
1337 | * error code? */ | |
f1588b1f | 1338 | dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, packet); |
064af421 BP |
1339 | return 0; |
1340 | } | |
1341 | ||
fa8b054f BP |
1342 | /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and |
1343 | * performs the 'n_actions' actions in 'actions'. The new flow will not | |
1344 | * timeout. | |
1345 | * | |
1346 | * If cls_rule->priority is in the range of priorities supported by OpenFlow | |
1347 | * (0...65535, inclusive) then the flow will be visible to OpenFlow | |
1348 | * controllers; otherwise, it will be hidden. | |
1349 | * | |
1350 | * The caller retains ownership of 'cls_rule' and 'actions'. */ | |
064af421 | 1351 | void |
cf3fad8a | 1352 | ofproto_add_flow(struct ofproto *p, const struct cls_rule *cls_rule, |
fa8b054f | 1353 | const union ofp_action *actions, size_t n_actions) |
064af421 BP |
1354 | { |
1355 | struct rule *rule; | |
bcf84111 | 1356 | rule = rule_create(cls_rule, actions, n_actions, 0, 0, 0, false); |
afe75089 | 1357 | rule_insert(p, rule); |
064af421 BP |
1358 | } |
1359 | ||
1360 | void | |
cf3fad8a | 1361 | ofproto_delete_flow(struct ofproto *ofproto, const struct cls_rule *target) |
064af421 BP |
1362 | { |
1363 | struct rule *rule; | |
1364 | ||
1365 | rule = rule_from_cls_rule(classifier_find_rule_exactly(&ofproto->cls, | |
cf3fad8a | 1366 | target)); |
064af421 BP |
1367 | if (rule) { |
1368 | rule_remove(ofproto, rule); | |
1369 | } | |
1370 | } | |
1371 | ||
064af421 BP |
1372 | void |
1373 | ofproto_flush_flows(struct ofproto *ofproto) | |
1374 | { | |
bcf84111 | 1375 | struct facet *facet, *next_facet; |
5ecc9d81 BP |
1376 | struct rule *rule, *next_rule; |
1377 | struct cls_cursor cursor; | |
bcf84111 | 1378 | |
064af421 | 1379 | COVERAGE_INC(ofproto_flush); |
bcf84111 BP |
1380 | |
1381 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1382 | /* Mark the facet as not installed so that facet_remove() doesn't | |
1383 | * bother trying to uninstall it. There is no point in uninstalling it | |
1384 | * individually since we are about to blow away all the facets with | |
1385 | * dpif_flow_flush(). */ | |
1386 | facet->installed = false; | |
1387 | facet_remove(ofproto, facet); | |
1388 | } | |
5ecc9d81 BP |
1389 | |
1390 | cls_cursor_init(&cursor, &ofproto->cls, NULL); | |
1391 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
1392 | rule_remove(ofproto, rule); | |
1393 | } | |
1394 | ||
c228a364 | 1395 | dpif_flow_flush(ofproto->dpif); |
064af421 BP |
1396 | if (ofproto->in_band) { |
1397 | in_band_flushed(ofproto->in_band); | |
1398 | } | |
1399 | if (ofproto->fail_open) { | |
1400 | fail_open_flushed(ofproto->fail_open); | |
1401 | } | |
1402 | } | |
1403 | \f | |
1404 | static void | |
1405 | reinit_ports(struct ofproto *p) | |
1406 | { | |
1407 | struct svec devnames; | |
1408 | struct ofport *ofport; | |
064af421 BP |
1409 | struct odp_port *odp_ports; |
1410 | size_t n_odp_ports; | |
1411 | size_t i; | |
1412 | ||
898bf89d JP |
1413 | COVERAGE_INC(ofproto_reinit_ports); |
1414 | ||
064af421 | 1415 | svec_init(&devnames); |
4e8e4213 | 1416 | HMAP_FOR_EACH (ofport, hmap_node, &p->ports) { |
064af421 BP |
1417 | svec_add (&devnames, (char *) ofport->opp.name); |
1418 | } | |
c228a364 | 1419 | dpif_port_list(p->dpif, &odp_ports, &n_odp_ports); |
064af421 BP |
1420 | for (i = 0; i < n_odp_ports; i++) { |
1421 | svec_add (&devnames, odp_ports[i].devname); | |
1422 | } | |
1423 | free(odp_ports); | |
1424 | ||
1425 | svec_sort_unique(&devnames); | |
1426 | for (i = 0; i < devnames.n; i++) { | |
1427 | update_port(p, devnames.names[i]); | |
1428 | } | |
1429 | svec_destroy(&devnames); | |
1430 | } | |
1431 | ||
064af421 BP |
1432 | static struct ofport * |
1433 | make_ofport(const struct odp_port *odp_port) | |
1434 | { | |
149f577a | 1435 | struct netdev_options netdev_options; |
064af421 BP |
1436 | enum netdev_flags flags; |
1437 | struct ofport *ofport; | |
1438 | struct netdev *netdev; | |
064af421 BP |
1439 | int error; |
1440 | ||
149f577a JG |
1441 | memset(&netdev_options, 0, sizeof netdev_options); |
1442 | netdev_options.name = odp_port->devname; | |
1443 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
149f577a JG |
1444 | |
1445 | error = netdev_open(&netdev_options, &netdev); | |
064af421 BP |
1446 | if (error) { |
1447 | VLOG_WARN_RL(&rl, "ignoring port %s (%"PRIu16") because netdev %s " | |
1448 | "cannot be opened (%s)", | |
1449 | odp_port->devname, odp_port->port, | |
1450 | odp_port->devname, strerror(error)); | |
1451 | return NULL; | |
1452 | } | |
1453 | ||
1454 | ofport = xmalloc(sizeof *ofport); | |
1455 | ofport->netdev = netdev; | |
ca0f572c | 1456 | ofport->odp_port = odp_port->port; |
064af421 | 1457 | ofport->opp.port_no = odp_port_to_ofp_port(odp_port->port); |
80992a35 | 1458 | netdev_get_etheraddr(netdev, ofport->opp.hw_addr); |
064af421 BP |
1459 | memcpy(ofport->opp.name, odp_port->devname, |
1460 | MIN(sizeof ofport->opp.name, sizeof odp_port->devname)); | |
1461 | ofport->opp.name[sizeof ofport->opp.name - 1] = '\0'; | |
1462 | ||
1463 | netdev_get_flags(netdev, &flags); | |
1464 | ofport->opp.config = flags & NETDEV_UP ? 0 : OFPPC_PORT_DOWN; | |
1465 | ||
85da620e | 1466 | ofport->opp.state = netdev_get_carrier(netdev) ? 0 : OFPPS_LINK_DOWN; |
064af421 BP |
1467 | |
1468 | netdev_get_features(netdev, | |
1469 | &ofport->opp.curr, &ofport->opp.advertised, | |
1470 | &ofport->opp.supported, &ofport->opp.peer); | |
1471 | return ofport; | |
1472 | } | |
1473 | ||
1474 | static bool | |
1475 | ofport_conflicts(const struct ofproto *p, const struct odp_port *odp_port) | |
1476 | { | |
ca0f572c | 1477 | if (get_port(p, odp_port->port)) { |
064af421 BP |
1478 | VLOG_WARN_RL(&rl, "ignoring duplicate port %"PRIu16" in datapath", |
1479 | odp_port->port); | |
1480 | return true; | |
1481 | } else if (shash_find(&p->port_by_name, odp_port->devname)) { | |
1482 | VLOG_WARN_RL(&rl, "ignoring duplicate device %s in datapath", | |
1483 | odp_port->devname); | |
1484 | return true; | |
1485 | } else { | |
1486 | return false; | |
1487 | } | |
1488 | } | |
1489 | ||
1490 | static int | |
1491 | ofport_equal(const struct ofport *a_, const struct ofport *b_) | |
1492 | { | |
1493 | const struct ofp_phy_port *a = &a_->opp; | |
1494 | const struct ofp_phy_port *b = &b_->opp; | |
1495 | ||
1496 | BUILD_ASSERT_DECL(sizeof *a == 48); /* Detect ofp_phy_port changes. */ | |
1497 | return (a->port_no == b->port_no | |
1498 | && !memcmp(a->hw_addr, b->hw_addr, sizeof a->hw_addr) | |
1499 | && !strcmp((char *) a->name, (char *) b->name) | |
1500 | && a->state == b->state | |
1501 | && a->config == b->config | |
1502 | && a->curr == b->curr | |
1503 | && a->advertised == b->advertised | |
1504 | && a->supported == b->supported | |
1505 | && a->peer == b->peer); | |
1506 | } | |
1507 | ||
1508 | static void | |
1509 | send_port_status(struct ofproto *p, const struct ofport *ofport, | |
1510 | uint8_t reason) | |
1511 | { | |
1512 | /* XXX Should limit the number of queued port status change messages. */ | |
1513 | struct ofconn *ofconn; | |
4e8e4213 | 1514 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
064af421 BP |
1515 | struct ofp_port_status *ops; |
1516 | struct ofpbuf *b; | |
1517 | ||
197a992f BP |
1518 | /* Primary controllers, even slaves, should always get port status |
1519 | updates. Otherwise obey ofconn_receives_async_msgs(). */ | |
1520 | if (ofconn->type != OFCONN_PRIMARY | |
1521 | && !ofconn_receives_async_msgs(ofconn)) { | |
9deba63b BP |
1522 | continue; |
1523 | } | |
1524 | ||
064af421 BP |
1525 | ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &b); |
1526 | ops->reason = reason; | |
1527 | ops->desc = ofport->opp; | |
1528 | hton_ofp_phy_port(&ops->desc); | |
1529 | queue_tx(b, ofconn, NULL); | |
1530 | } | |
064af421 BP |
1531 | } |
1532 | ||
1533 | static void | |
1534 | ofport_install(struct ofproto *p, struct ofport *ofport) | |
1535 | { | |
72b06300 BP |
1536 | const char *netdev_name = (const char *) ofport->opp.name; |
1537 | ||
e9e28be3 | 1538 | netdev_monitor_add(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1539 | hmap_insert(&p->ports, &ofport->hmap_node, hash_int(ofport->odp_port, 0)); |
72b06300 BP |
1540 | shash_add(&p->port_by_name, netdev_name, ofport); |
1541 | if (p->sflow) { | |
ca0f572c | 1542 | ofproto_sflow_add_port(p->sflow, ofport->odp_port, netdev_name); |
72b06300 | 1543 | } |
064af421 BP |
1544 | } |
1545 | ||
1546 | static void | |
1547 | ofport_remove(struct ofproto *p, struct ofport *ofport) | |
1548 | { | |
e9e28be3 | 1549 | netdev_monitor_remove(p->netdev_monitor, ofport->netdev); |
ca0f572c | 1550 | hmap_remove(&p->ports, &ofport->hmap_node); |
064af421 BP |
1551 | shash_delete(&p->port_by_name, |
1552 | shash_find(&p->port_by_name, (char *) ofport->opp.name)); | |
72b06300 | 1553 | if (p->sflow) { |
ca0f572c | 1554 | ofproto_sflow_del_port(p->sflow, ofport->odp_port); |
72b06300 | 1555 | } |
064af421 BP |
1556 | } |
1557 | ||
1558 | static void | |
1559 | ofport_free(struct ofport *ofport) | |
1560 | { | |
1561 | if (ofport) { | |
1562 | netdev_close(ofport->netdev); | |
1563 | free(ofport); | |
1564 | } | |
1565 | } | |
1566 | ||
ca0f572c BP |
1567 | static struct ofport * |
1568 | get_port(const struct ofproto *ofproto, uint16_t odp_port) | |
1569 | { | |
1570 | struct ofport *port; | |
1571 | ||
4e8e4213 | 1572 | HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, |
ca0f572c BP |
1573 | hash_int(odp_port, 0), &ofproto->ports) { |
1574 | if (port->odp_port == odp_port) { | |
1575 | return port; | |
1576 | } | |
1577 | } | |
1578 | return NULL; | |
1579 | } | |
1580 | ||
064af421 BP |
1581 | static void |
1582 | update_port(struct ofproto *p, const char *devname) | |
1583 | { | |
1584 | struct odp_port odp_port; | |
c874dc6d BP |
1585 | struct ofport *old_ofport; |
1586 | struct ofport *new_ofport; | |
064af421 BP |
1587 | int error; |
1588 | ||
1589 | COVERAGE_INC(ofproto_update_port); | |
c874dc6d BP |
1590 | |
1591 | /* Query the datapath for port information. */ | |
c228a364 | 1592 | error = dpif_port_query_by_name(p->dpif, devname, &odp_port); |
064af421 | 1593 | |
c874dc6d BP |
1594 | /* Find the old ofport. */ |
1595 | old_ofport = shash_find_data(&p->port_by_name, devname); | |
1596 | if (!error) { | |
1597 | if (!old_ofport) { | |
1598 | /* There's no port named 'devname' but there might be a port with | |
1599 | * the same port number. This could happen if a port is deleted | |
1600 | * and then a new one added in its place very quickly, or if a port | |
1601 | * is renamed. In the former case we want to send an OFPPR_DELETE | |
1602 | * and an OFPPR_ADD, and in the latter case we want to send a | |
1603 | * single OFPPR_MODIFY. We can distinguish the cases by comparing | |
1604 | * the old port's ifindex against the new port, or perhaps less | |
1605 | * reliably but more portably by comparing the old port's MAC | |
1606 | * against the new port's MAC. However, this code isn't that smart | |
1607 | * and always sends an OFPPR_MODIFY (XXX). */ | |
ca0f572c | 1608 | old_ofport = get_port(p, odp_port.port); |
064af421 | 1609 | } |
c874dc6d | 1610 | } else if (error != ENOENT && error != ENODEV) { |
064af421 BP |
1611 | VLOG_WARN_RL(&rl, "dpif_port_query_by_name returned unexpected error " |
1612 | "%s", strerror(error)); | |
1613 | return; | |
1614 | } | |
c874dc6d BP |
1615 | |
1616 | /* Create a new ofport. */ | |
1617 | new_ofport = !error ? make_ofport(&odp_port) : NULL; | |
1618 | ||
1619 | /* Eliminate a few pathological cases. */ | |
1620 | if (!old_ofport && !new_ofport) { | |
1621 | return; | |
1622 | } else if (old_ofport && new_ofport) { | |
1623 | /* Most of the 'config' bits are OpenFlow soft state, but | |
bc4a55cd BP |
1624 | * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the |
1625 | * OpenFlow bits from old_ofport. (make_ofport() only sets | |
1626 | * OFPPC_PORT_DOWN and leaves the other bits 0.) */ | |
c874dc6d BP |
1627 | new_ofport->opp.config |= old_ofport->opp.config & ~OFPPC_PORT_DOWN; |
1628 | ||
1629 | if (ofport_equal(old_ofport, new_ofport)) { | |
1630 | /* False alarm--no change. */ | |
1631 | ofport_free(new_ofport); | |
1632 | return; | |
1633 | } | |
1634 | } | |
1635 | ||
1636 | /* Now deal with the normal cases. */ | |
1637 | if (old_ofport) { | |
1638 | ofport_remove(p, old_ofport); | |
1639 | } | |
1640 | if (new_ofport) { | |
1641 | ofport_install(p, new_ofport); | |
1642 | } | |
1643 | send_port_status(p, new_ofport ? new_ofport : old_ofport, | |
1644 | (!old_ofport ? OFPPR_ADD | |
1645 | : !new_ofport ? OFPPR_DELETE | |
1646 | : OFPPR_MODIFY)); | |
1647 | ofport_free(old_ofport); | |
064af421 BP |
1648 | } |
1649 | ||
1650 | static int | |
1651 | init_ports(struct ofproto *p) | |
1652 | { | |
1653 | struct odp_port *ports; | |
1654 | size_t n_ports; | |
1655 | size_t i; | |
1656 | int error; | |
1657 | ||
c228a364 | 1658 | error = dpif_port_list(p->dpif, &ports, &n_ports); |
064af421 BP |
1659 | if (error) { |
1660 | return error; | |
1661 | } | |
1662 | ||
1663 | for (i = 0; i < n_ports; i++) { | |
1664 | const struct odp_port *odp_port = &ports[i]; | |
1665 | if (!ofport_conflicts(p, odp_port)) { | |
1666 | struct ofport *ofport = make_ofport(odp_port); | |
1667 | if (ofport) { | |
1668 | ofport_install(p, ofport); | |
1669 | } | |
1670 | } | |
1671 | } | |
1672 | free(ports); | |
064af421 BP |
1673 | return 0; |
1674 | } | |
1675 | \f | |
1676 | static struct ofconn * | |
76ce9432 | 1677 | ofconn_create(struct ofproto *p, struct rconn *rconn, enum ofconn_type type) |
064af421 | 1678 | { |
76ce9432 BP |
1679 | struct ofconn *ofconn = xzalloc(sizeof *ofconn); |
1680 | ofconn->ofproto = p; | |
064af421 BP |
1681 | list_push_back(&p->all_conns, &ofconn->node); |
1682 | ofconn->rconn = rconn; | |
76ce9432 | 1683 | ofconn->type = type; |
b70eac89 | 1684 | ofconn->flow_format = NXFF_OPENFLOW10; |
9deba63b | 1685 | ofconn->role = NX_ROLE_OTHER; |
76ce9432 | 1686 | ofconn->packet_in_counter = rconn_packet_counter_create (); |
064af421 | 1687 | ofconn->pktbuf = NULL; |
064af421 | 1688 | ofconn->miss_send_len = 0; |
064af421 BP |
1689 | ofconn->reply_counter = rconn_packet_counter_create (); |
1690 | return ofconn; | |
1691 | } | |
1692 | ||
1693 | static void | |
c475ae67 | 1694 | ofconn_destroy(struct ofconn *ofconn) |
064af421 | 1695 | { |
5899143f | 1696 | if (ofconn->type == OFCONN_PRIMARY) { |
76ce9432 BP |
1697 | hmap_remove(&ofconn->ofproto->controllers, &ofconn->hmap_node); |
1698 | } | |
1699 | discovery_destroy(ofconn->discovery); | |
1700 | ||
064af421 | 1701 | list_remove(&ofconn->node); |
76ce9432 | 1702 | switch_status_unregister(ofconn->ss); |
064af421 BP |
1703 | rconn_destroy(ofconn->rconn); |
1704 | rconn_packet_counter_destroy(ofconn->packet_in_counter); | |
1705 | rconn_packet_counter_destroy(ofconn->reply_counter); | |
1706 | pktbuf_destroy(ofconn->pktbuf); | |
1707 | free(ofconn); | |
1708 | } | |
1709 | ||
1710 | static void | |
3269c562 | 1711 | ofconn_run(struct ofconn *ofconn) |
064af421 | 1712 | { |
3269c562 | 1713 | struct ofproto *p = ofconn->ofproto; |
064af421 | 1714 | int iteration; |
76ce9432 BP |
1715 | size_t i; |
1716 | ||
1717 | if (ofconn->discovery) { | |
1718 | char *controller_name; | |
1719 | if (rconn_is_connectivity_questionable(ofconn->rconn)) { | |
1720 | discovery_question_connectivity(ofconn->discovery); | |
1721 | } | |
1722 | if (discovery_run(ofconn->discovery, &controller_name)) { | |
1723 | if (controller_name) { | |
eb15cdbb BP |
1724 | char *ofconn_name = ofconn_make_name(p, controller_name); |
1725 | rconn_connect(ofconn->rconn, controller_name, ofconn_name); | |
1726 | free(ofconn_name); | |
76ce9432 BP |
1727 | } else { |
1728 | rconn_disconnect(ofconn->rconn); | |
1729 | } | |
1730 | } | |
1731 | } | |
1732 | ||
1733 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1734 | pinsched_run(ofconn->schedulers[i], do_send_packet_in, ofconn); | |
1735 | } | |
064af421 BP |
1736 | |
1737 | rconn_run(ofconn->rconn); | |
1738 | ||
1739 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1740 | /* Limit the number of iterations to prevent other tasks from | |
1741 | * starving. */ | |
1742 | for (iteration = 0; iteration < 50; iteration++) { | |
1743 | struct ofpbuf *of_msg = rconn_recv(ofconn->rconn); | |
1744 | if (!of_msg) { | |
1745 | break; | |
1746 | } | |
7778bd15 BP |
1747 | if (p->fail_open) { |
1748 | fail_open_maybe_recover(p->fail_open); | |
1749 | } | |
3269c562 | 1750 | handle_openflow(ofconn, of_msg); |
064af421 BP |
1751 | ofpbuf_delete(of_msg); |
1752 | } | |
1753 | } | |
1754 | ||
76ce9432 | 1755 | if (!ofconn->discovery && !rconn_is_alive(ofconn->rconn)) { |
c475ae67 | 1756 | ofconn_destroy(ofconn); |
064af421 BP |
1757 | } |
1758 | } | |
1759 | ||
1760 | static void | |
1761 | ofconn_wait(struct ofconn *ofconn) | |
1762 | { | |
76ce9432 BP |
1763 | int i; |
1764 | ||
1765 | if (ofconn->discovery) { | |
1766 | discovery_wait(ofconn->discovery); | |
1767 | } | |
1768 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1769 | pinsched_wait(ofconn->schedulers[i]); | |
1770 | } | |
064af421 BP |
1771 | rconn_run_wait(ofconn->rconn); |
1772 | if (rconn_packet_counter_read (ofconn->reply_counter) < OFCONN_REPLY_MAX) { | |
1773 | rconn_recv_wait(ofconn->rconn); | |
1774 | } else { | |
1775 | COVERAGE_INC(ofproto_ofconn_stuck); | |
1776 | } | |
1777 | } | |
c91248b3 BP |
1778 | |
1779 | /* Returns true if 'ofconn' should receive asynchronous messages. */ | |
1780 | static bool | |
1781 | ofconn_receives_async_msgs(const struct ofconn *ofconn) | |
1782 | { | |
5899143f BP |
1783 | if (ofconn->type == OFCONN_PRIMARY) { |
1784 | /* Primary controllers always get asynchronous messages unless they | |
c91248b3 BP |
1785 | * have configured themselves as "slaves". */ |
1786 | return ofconn->role != NX_ROLE_SLAVE; | |
1787 | } else { | |
5899143f BP |
1788 | /* Service connections don't get asynchronous messages unless they have |
1789 | * explicitly asked for them by setting a nonzero miss send length. */ | |
c91248b3 BP |
1790 | return ofconn->miss_send_len > 0; |
1791 | } | |
1792 | } | |
eb15cdbb BP |
1793 | |
1794 | /* Returns a human-readable name for an OpenFlow connection between 'ofproto' | |
1795 | * and 'target', suitable for use in log messages for identifying the | |
1796 | * connection. | |
1797 | * | |
1798 | * The name is dynamically allocated. The caller should free it (with free()) | |
1799 | * when it is no longer needed. */ | |
1800 | static char * | |
1801 | ofconn_make_name(const struct ofproto *ofproto, const char *target) | |
1802 | { | |
1803 | return xasprintf("%s<->%s", dpif_base_name(ofproto->dpif), target); | |
1804 | } | |
7d674866 BP |
1805 | |
1806 | static void | |
1807 | ofconn_set_rate_limit(struct ofconn *ofconn, int rate, int burst) | |
1808 | { | |
1809 | int i; | |
1810 | ||
1811 | for (i = 0; i < N_SCHEDULERS; i++) { | |
1812 | struct pinsched **s = &ofconn->schedulers[i]; | |
1813 | ||
1814 | if (rate > 0) { | |
1815 | if (!*s) { | |
1816 | *s = pinsched_create(rate, burst, | |
1817 | ofconn->ofproto->switch_status); | |
1818 | } else { | |
1819 | pinsched_set_limits(*s, rate, burst); | |
1820 | } | |
1821 | } else { | |
1822 | pinsched_destroy(*s); | |
1823 | *s = NULL; | |
1824 | } | |
1825 | } | |
1826 | } | |
1827 | \f | |
1828 | static void | |
1829 | ofservice_reconfigure(struct ofservice *ofservice, | |
1830 | const struct ofproto_controller *c) | |
1831 | { | |
1832 | ofservice->probe_interval = c->probe_interval; | |
1833 | ofservice->rate_limit = c->rate_limit; | |
1834 | ofservice->burst_limit = c->burst_limit; | |
1835 | } | |
1836 | ||
1837 | /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a | |
1838 | * positive errno value. */ | |
1839 | static int | |
1840 | ofservice_create(struct ofproto *ofproto, const struct ofproto_controller *c) | |
1841 | { | |
1842 | struct ofservice *ofservice; | |
1843 | struct pvconn *pvconn; | |
1844 | int error; | |
1845 | ||
1846 | error = pvconn_open(c->target, &pvconn); | |
1847 | if (error) { | |
1848 | return error; | |
1849 | } | |
1850 | ||
1851 | ofservice = xzalloc(sizeof *ofservice); | |
1852 | hmap_insert(&ofproto->services, &ofservice->node, | |
1853 | hash_string(c->target, 0)); | |
1854 | ofservice->pvconn = pvconn; | |
1855 | ||
1856 | ofservice_reconfigure(ofservice, c); | |
1857 | ||
1858 | return 0; | |
1859 | } | |
1860 | ||
1861 | static void | |
1862 | ofservice_destroy(struct ofproto *ofproto, struct ofservice *ofservice) | |
1863 | { | |
1864 | hmap_remove(&ofproto->services, &ofservice->node); | |
1865 | pvconn_close(ofservice->pvconn); | |
1866 | free(ofservice); | |
1867 | } | |
1868 | ||
1869 | /* Finds and returns the ofservice within 'ofproto' that has the given | |
1870 | * 'target', or a null pointer if none exists. */ | |
1871 | static struct ofservice * | |
1872 | ofservice_lookup(struct ofproto *ofproto, const char *target) | |
1873 | { | |
1874 | struct ofservice *ofservice; | |
1875 | ||
4e8e4213 BP |
1876 | HMAP_FOR_EACH_WITH_HASH (ofservice, node, hash_string(target, 0), |
1877 | &ofproto->services) { | |
7d674866 BP |
1878 | if (!strcmp(pvconn_get_name(ofservice->pvconn), target)) { |
1879 | return ofservice; | |
1880 | } | |
1881 | } | |
1882 | return NULL; | |
1883 | } | |
064af421 | 1884 | \f |
bcf84111 BP |
1885 | /* Returns true if 'rule' should be hidden from the controller. |
1886 | * | |
1887 | * Rules with priority higher than UINT16_MAX are set up by ofproto itself | |
1888 | * (e.g. by in-band control) and are intentionally hidden from the | |
1889 | * controller. */ | |
1890 | static bool | |
1891 | rule_is_hidden(const struct rule *rule) | |
1892 | { | |
1893 | return rule->cr.priority > UINT16_MAX; | |
1894 | } | |
1895 | ||
1896 | /* Creates and returns a new rule initialized as specified. | |
1897 | * | |
1898 | * The caller is responsible for inserting the rule into the classifier (with | |
1899 | * rule_insert()). */ | |
064af421 | 1900 | static struct rule * |
bcf84111 | 1901 | rule_create(const struct cls_rule *cls_rule, |
064af421 | 1902 | const union ofp_action *actions, size_t n_actions, |
ca069229 | 1903 | uint16_t idle_timeout, uint16_t hard_timeout, |
8054fc48 | 1904 | ovs_be64 flow_cookie, bool send_flow_removed) |
064af421 | 1905 | { |
ec6fde61 | 1906 | struct rule *rule = xzalloc(sizeof *rule); |
bcf84111 | 1907 | rule->cr = *cls_rule; |
064af421 BP |
1908 | rule->idle_timeout = idle_timeout; |
1909 | rule->hard_timeout = hard_timeout; | |
39997502 | 1910 | rule->flow_cookie = flow_cookie; |
064af421 | 1911 | rule->used = rule->created = time_msec(); |
ca069229 | 1912 | rule->send_flow_removed = send_flow_removed; |
bcf84111 | 1913 | list_init(&rule->facets); |
3dffcf07 BP |
1914 | if (n_actions > 0) { |
1915 | rule->n_actions = n_actions; | |
1916 | rule->actions = xmemdup(actions, n_actions * sizeof *actions); | |
1917 | } | |
0193b2af | 1918 | |
064af421 BP |
1919 | return rule; |
1920 | } | |
1921 | ||
1922 | static struct rule * | |
1923 | rule_from_cls_rule(const struct cls_rule *cls_rule) | |
1924 | { | |
1925 | return cls_rule ? CONTAINER_OF(cls_rule, struct rule, cr) : NULL; | |
1926 | } | |
1927 | ||
1928 | static void | |
1929 | rule_free(struct rule *rule) | |
1930 | { | |
1931 | free(rule->actions); | |
064af421 BP |
1932 | free(rule); |
1933 | } | |
1934 | ||
bcf84111 BP |
1935 | /* Destroys 'rule' and iterates through all of its facets and revalidates them, |
1936 | * destroying any that no longer has a rule (which is probably all of them). | |
064af421 | 1937 | * |
bcf84111 | 1938 | * The caller must have already removed 'rule' from the classifier. */ |
064af421 BP |
1939 | static void |
1940 | rule_destroy(struct ofproto *ofproto, struct rule *rule) | |
1941 | { | |
bcf84111 BP |
1942 | struct facet *facet, *next_facet; |
1943 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
1944 | facet_revalidate(ofproto, facet); | |
064af421 BP |
1945 | } |
1946 | rule_free(rule); | |
1947 | } | |
1948 | ||
bcf84111 BP |
1949 | /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action |
1950 | * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't | |
1951 | * count). */ | |
064af421 | 1952 | static bool |
8054fc48 | 1953 | rule_has_out_port(const struct rule *rule, ovs_be16 out_port) |
064af421 BP |
1954 | { |
1955 | const union ofp_action *oa; | |
1956 | struct actions_iterator i; | |
1957 | ||
1958 | if (out_port == htons(OFPP_NONE)) { | |
1959 | return true; | |
1960 | } | |
1961 | for (oa = actions_first(&i, rule->actions, rule->n_actions); oa; | |
1962 | oa = actions_next(&i)) { | |
c1c9c9c4 | 1963 | if (action_outputs_to_port(oa, out_port)) { |
064af421 BP |
1964 | return true; |
1965 | } | |
1966 | } | |
1967 | return false; | |
1968 | } | |
1969 | ||
750638bb BP |
1970 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on |
1971 | * 'packet', which arrived on 'in_port'. | |
1972 | * | |
1973 | * Takes ownership of 'packet'. */ | |
9dbb9d5e BP |
1974 | static bool |
1975 | execute_odp_actions(struct ofproto *ofproto, uint16_t in_port, | |
1976 | const union odp_action *actions, size_t n_actions, | |
750638bb | 1977 | struct ofpbuf *packet) |
9dbb9d5e | 1978 | { |
750638bb | 1979 | if (n_actions == 1 && actions[0].type == ODPAT_CONTROLLER) { |
9dbb9d5e BP |
1980 | /* As an optimization, avoid a round-trip from userspace to kernel to |
1981 | * userspace. This also avoids possibly filling up kernel packet | |
1982 | * buffers along the way. */ | |
9dbb9d5e BP |
1983 | struct odp_msg *msg; |
1984 | ||
750638bb | 1985 | msg = ofpbuf_push_uninit(packet, sizeof *msg); |
9dbb9d5e BP |
1986 | msg->type = _ODPL_ACTION_NR; |
1987 | msg->length = sizeof(struct odp_msg) + packet->size; | |
1988 | msg->port = in_port; | |
1989 | msg->reserved = 0; | |
1990 | msg->arg = actions[0].controller.arg; | |
9dbb9d5e | 1991 | |
750638bb | 1992 | send_packet_in(ofproto, packet); |
9dbb9d5e | 1993 | |
750638bb BP |
1994 | return true; |
1995 | } else { | |
1996 | int error; | |
9dbb9d5e | 1997 | |
f1588b1f | 1998 | error = dpif_execute(ofproto->dpif, actions, n_actions, packet); |
750638bb BP |
1999 | ofpbuf_delete(packet); |
2000 | return !error; | |
2001 | } | |
9dbb9d5e BP |
2002 | } |
2003 | ||
bcf84111 BP |
2004 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s |
2005 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2006 | * ofp_packet_in) bytes of headroom. | |
064af421 | 2007 | * |
bcf84111 BP |
2008 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, |
2009 | * applying flow_extract() to 'packet' would yield the same flow as | |
2010 | * 'facet->flow'. | |
064af421 | 2011 | * |
bcf84111 BP |
2012 | * 'facet' must have accurately composed ODP actions; that is, it must not be |
2013 | * in need of revalidation. | |
750638bb BP |
2014 | * |
2015 | * Takes ownership of 'packet'. */ | |
064af421 | 2016 | static void |
bcf84111 BP |
2017 | facet_execute(struct ofproto *ofproto, struct facet *facet, |
2018 | struct ofpbuf *packet) | |
064af421 | 2019 | { |
750638bb | 2020 | struct odp_flow_stats stats; |
bcf84111 BP |
2021 | |
2022 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2023 | ||
2024 | flow_extract_stats(&facet->flow, packet, &stats); | |
2025 | if (execute_odp_actions(ofproto, facet->flow.in_port, | |
2026 | facet->actions, facet->n_actions, packet)) { | |
2027 | facet_update_stats(ofproto, facet, &stats); | |
2028 | facet->used = time_msec(); | |
2029 | netflow_flow_update_time(ofproto->netflow, | |
2030 | &facet->nf_flow, facet->used); | |
2031 | } | |
2032 | } | |
2033 | ||
2034 | /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s | |
2035 | * statistics (or the statistics for one of its facets) appropriately. | |
2036 | * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom. | |
2037 | * | |
2038 | * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited | |
2039 | * with statistics for 'packet' either way. | |
2040 | * | |
2041 | * Takes ownership of 'packet'. */ | |
2042 | static void | |
2043 | rule_execute(struct ofproto *ofproto, struct rule *rule, uint16_t in_port, | |
2044 | struct ofpbuf *packet) | |
2045 | { | |
2046 | struct facet *facet; | |
064af421 | 2047 | struct odp_actions a; |
bcf84111 BP |
2048 | struct flow flow; |
2049 | size_t size; | |
064af421 | 2050 | |
750638bb BP |
2051 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); |
2052 | ||
bcf84111 BP |
2053 | flow_extract(packet, 0, in_port, &flow); |
2054 | ||
2055 | /* First look for a related facet. If we find one, account it to that. */ | |
2056 | facet = facet_lookup_valid(ofproto, &flow); | |
2057 | if (facet && facet->rule == rule) { | |
2058 | facet_execute(ofproto, facet, packet); | |
2059 | return; | |
064af421 BP |
2060 | } |
2061 | ||
bcf84111 BP |
2062 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then |
2063 | * create a new facet for it and use that. */ | |
2064 | if (rule_lookup(ofproto, &flow) == rule) { | |
2065 | facet = facet_create(ofproto, rule, &flow, packet); | |
2066 | facet_execute(ofproto, facet, packet); | |
2067 | facet_install(ofproto, facet, true); | |
2068 | return; | |
2069 | } | |
2070 | ||
2071 | /* We can't account anything to a facet. If we were to try, then that | |
2072 | * facet would have a non-matching rule, busting our invariants. */ | |
2073 | if (xlate_actions(rule->actions, rule->n_actions, &flow, ofproto, | |
2074 | packet, &a, NULL, 0, NULL)) { | |
2075 | ofpbuf_delete(packet); | |
2076 | return; | |
2077 | } | |
2078 | size = packet->size; | |
2079 | if (execute_odp_actions(ofproto, in_port, | |
2080 | a.actions, a.n_actions, packet)) { | |
064af421 | 2081 | rule->used = time_msec(); |
bcf84111 BP |
2082 | rule->packet_count++; |
2083 | rule->byte_count += size; | |
064af421 BP |
2084 | } |
2085 | } | |
2086 | ||
afe75089 | 2087 | /* Inserts 'rule' into 'p''s flow table. */ |
064af421 | 2088 | static void |
afe75089 | 2089 | rule_insert(struct ofproto *p, struct rule *rule) |
064af421 BP |
2090 | { |
2091 | struct rule *displaced_rule; | |
2092 | ||
064af421 | 2093 | displaced_rule = rule_from_cls_rule(classifier_insert(&p->cls, &rule->cr)); |
bcf84111 BP |
2094 | if (displaced_rule) { |
2095 | rule_destroy(p, displaced_rule); | |
064af421 | 2096 | } |
bcf84111 | 2097 | p->need_revalidate = true; |
064af421 BP |
2098 | } |
2099 | ||
bcf84111 BP |
2100 | /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a |
2101 | * 'flow' and an example 'packet' within that flow. | |
2102 | * | |
2103 | * The caller must already have determined that no facet with an identical | |
2104 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
2105 | * 'ofproto''s classifier table. */ | |
2106 | static struct facet * | |
2107 | facet_create(struct ofproto *ofproto, struct rule *rule, | |
2108 | const struct flow *flow, const struct ofpbuf *packet) | |
064af421 | 2109 | { |
bcf84111 | 2110 | struct facet *facet; |
fbb2ea0b | 2111 | |
bcf84111 BP |
2112 | facet = xzalloc(sizeof *facet); |
2113 | facet->used = time_msec(); | |
2114 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2115 | list_push_back(&rule->facets, &facet->list_node); | |
2116 | facet->rule = rule; | |
2117 | facet->flow = *flow; | |
2118 | netflow_flow_init(&facet->nf_flow); | |
2119 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2120 | ||
2121 | facet_make_actions(ofproto, facet, packet); | |
064af421 | 2122 | |
bcf84111 BP |
2123 | return facet; |
2124 | } | |
2125 | ||
2126 | static void | |
2127 | facet_free(struct facet *facet) | |
2128 | { | |
2129 | free(facet->actions); | |
2130 | free(facet); | |
064af421 BP |
2131 | } |
2132 | ||
431d4707 | 2133 | /* Remove 'rule' from 'ofproto' and free up the associated memory: |
431d4707 BP |
2134 | * |
2135 | * - Removes 'rule' from the classifier. | |
2136 | * | |
bcf84111 BP |
2137 | * - If 'rule' has facets, revalidates them (and possibly uninstalls and |
2138 | * destroys them), via rule_destroy(). | |
431d4707 | 2139 | */ |
064af421 BP |
2140 | static void |
2141 | rule_remove(struct ofproto *ofproto, struct rule *rule) | |
2142 | { | |
bcf84111 BP |
2143 | COVERAGE_INC(ofproto_del_rule); |
2144 | ofproto->need_revalidate = true; | |
064af421 BP |
2145 | classifier_remove(&ofproto->cls, &rule->cr); |
2146 | rule_destroy(ofproto, rule); | |
2147 | } | |
2148 | ||
bcf84111 BP |
2149 | /* Remove 'facet' from 'ofproto' and free up the associated memory: |
2150 | * | |
2151 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2152 | * rule's statistics, via facet_uninstall(). | |
2153 | * | |
2154 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2155 | */ | |
2156 | static void | |
2157 | facet_remove(struct ofproto *ofproto, struct facet *facet) | |
2158 | { | |
2159 | facet_uninstall(ofproto, facet); | |
d530fcd2 | 2160 | facet_flush_stats(ofproto, facet); |
bcf84111 BP |
2161 | hmap_remove(&ofproto->facets, &facet->hmap_node); |
2162 | list_remove(&facet->list_node); | |
2163 | facet_free(facet); | |
2164 | } | |
2165 | ||
7f7ae89d BP |
2166 | /* Composes the ODP actions for 'facet' based on its rule's actions. */ |
2167 | static void | |
bcf84111 BP |
2168 | facet_make_actions(struct ofproto *p, struct facet *facet, |
2169 | const struct ofpbuf *packet) | |
064af421 | 2170 | { |
bcf84111 | 2171 | const struct rule *rule = facet->rule; |
064af421 BP |
2172 | struct odp_actions a; |
2173 | size_t actions_len; | |
2174 | ||
bcf84111 BP |
2175 | xlate_actions(rule->actions, rule->n_actions, &facet->flow, p, |
2176 | packet, &a, &facet->tags, &facet->may_install, | |
2177 | &facet->nf_flow.output_iface); | |
064af421 BP |
2178 | |
2179 | actions_len = a.n_actions * sizeof *a.actions; | |
7f7ae89d BP |
2180 | if (facet->n_actions != a.n_actions |
2181 | || memcmp(facet->actions, a.actions, actions_len)) { | |
2182 | free(facet->actions); | |
2183 | facet->n_actions = a.n_actions; | |
2184 | facet->actions = xmemdup(a.actions, actions_len); | |
064af421 BP |
2185 | } |
2186 | } | |
2187 | ||
2188 | static int | |
bcf84111 | 2189 | facet_put__(struct ofproto *ofproto, struct facet *facet, int flags, |
064af421 BP |
2190 | struct odp_flow_put *put) |
2191 | { | |
2192 | memset(&put->flow.stats, 0, sizeof put->flow.stats); | |
bcf84111 BP |
2193 | odp_flow_key_from_flow(&put->flow.key, &facet->flow); |
2194 | put->flow.actions = facet->actions; | |
2195 | put->flow.n_actions = facet->n_actions; | |
ab48643b | 2196 | put->flow.flags = 0; |
064af421 | 2197 | put->flags = flags; |
c228a364 | 2198 | return dpif_flow_put(ofproto->dpif, put); |
064af421 BP |
2199 | } |
2200 | ||
bcf84111 BP |
2201 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If |
2202 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2203 | * 'facet'. */ | |
064af421 | 2204 | static void |
bcf84111 | 2205 | facet_install(struct ofproto *p, struct facet *facet, bool zero_stats) |
064af421 | 2206 | { |
bcf84111 | 2207 | if (facet->may_install) { |
064af421 | 2208 | struct odp_flow_put put; |
bcf84111 | 2209 | int flags; |
064af421 | 2210 | |
bcf84111 BP |
2211 | flags = ODPPF_CREATE | ODPPF_MODIFY; |
2212 | if (zero_stats) { | |
2213 | flags |= ODPPF_ZERO_STATS; | |
2214 | } | |
2215 | if (!facet_put__(p, facet, flags, &put)) { | |
2216 | facet->installed = true; | |
2217 | } | |
064af421 BP |
2218 | } |
2219 | } | |
2220 | ||
bcf84111 BP |
2221 | /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up |
2222 | * to the accounting hook function in the ofhooks structure. */ | |
064af421 | 2223 | static void |
bcf84111 BP |
2224 | facet_account(struct ofproto *ofproto, |
2225 | struct facet *facet, uint64_t extra_bytes) | |
064af421 | 2226 | { |
bcf84111 | 2227 | uint64_t total_bytes = facet->byte_count + extra_bytes; |
064af421 BP |
2228 | |
2229 | if (ofproto->ofhooks->account_flow_cb | |
bcf84111 | 2230 | && total_bytes > facet->accounted_bytes) |
064af421 BP |
2231 | { |
2232 | ofproto->ofhooks->account_flow_cb( | |
bcf84111 BP |
2233 | &facet->flow, facet->tags, facet->actions, facet->n_actions, |
2234 | total_bytes - facet->accounted_bytes, ofproto->aux); | |
2235 | facet->accounted_bytes = total_bytes; | |
064af421 BP |
2236 | } |
2237 | } | |
2238 | ||
d530fcd2 | 2239 | /* If 'rule' is installed in the datapath, uninstalls it. */ |
064af421 | 2240 | static void |
bcf84111 | 2241 | facet_uninstall(struct ofproto *p, struct facet *facet) |
064af421 | 2242 | { |
bcf84111 | 2243 | if (facet->installed) { |
064af421 BP |
2244 | struct odp_flow odp_flow; |
2245 | ||
bcf84111 | 2246 | odp_flow_key_from_flow(&odp_flow.key, &facet->flow); |
064af421 BP |
2247 | odp_flow.actions = NULL; |
2248 | odp_flow.n_actions = 0; | |
ab48643b | 2249 | odp_flow.flags = 0; |
c228a364 | 2250 | if (!dpif_flow_del(p->dpif, &odp_flow)) { |
bcf84111 | 2251 | facet_update_stats(p, facet, &odp_flow.stats); |
064af421 | 2252 | } |
bcf84111 | 2253 | facet->installed = false; |
064af421 BP |
2254 | } |
2255 | } | |
2256 | ||
bcf84111 BP |
2257 | /* Returns true if the only action for 'facet' is to send to the controller. |
2258 | * (We don't report NetFlow expiration messages for such facets because they | |
2259 | * are just part of the control logic for the network, not real traffic). */ | |
0193b2af | 2260 | static bool |
bcf84111 | 2261 | facet_is_controller_flow(struct facet *facet) |
0193b2af | 2262 | { |
bcf84111 BP |
2263 | return (facet |
2264 | && facet->rule->n_actions == 1 | |
2265 | && action_outputs_to_port(&facet->rule->actions[0], | |
c1c9c9c4 | 2266 | htons(OFPP_CONTROLLER))); |
0193b2af JG |
2267 | } |
2268 | ||
bcf84111 BP |
2269 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2270 | * accounting ofhook and emits a NetFlow expiration if appropriate. */ | |
064af421 | 2271 | static void |
d530fcd2 | 2272 | facet_flush_stats(struct ofproto *ofproto, struct facet *facet) |
064af421 | 2273 | { |
bcf84111 | 2274 | facet_account(ofproto, facet, 0); |
064af421 | 2275 | |
bcf84111 | 2276 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { |
064af421 | 2277 | struct ofexpired expired; |
bcf84111 BP |
2278 | expired.flow = facet->flow; |
2279 | expired.packet_count = facet->packet_count; | |
2280 | expired.byte_count = facet->byte_count; | |
2281 | expired.used = facet->used; | |
2282 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2283 | } | |
2284 | ||
2285 | facet->rule->packet_count += facet->packet_count; | |
2286 | facet->rule->byte_count += facet->byte_count; | |
2287 | ||
2288 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2289 | * reinstalled. */ | |
2290 | facet->packet_count = 0; | |
2291 | facet->byte_count = 0; | |
2292 | facet->accounted_bytes = 0; | |
2293 | ||
2294 | netflow_flow_clear(&facet->nf_flow); | |
2295 | } | |
2296 | ||
2297 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2298 | * Returns it if found, otherwise a null pointer. | |
2299 | * | |
2300 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2301 | * instead if that is important. */ | |
2302 | static struct facet * | |
2303 | facet_find(struct ofproto *ofproto, const struct flow *flow) | |
2304 | { | |
2305 | struct facet *facet; | |
2306 | ||
2307 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2308 | &ofproto->facets) { | |
2309 | if (flow_equal(flow, &facet->flow)) { | |
2310 | return facet; | |
2311 | } | |
2312 | } | |
2313 | ||
2314 | return NULL; | |
2315 | } | |
2316 | ||
2317 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2318 | * Returns it if found, otherwise a null pointer. | |
2319 | * | |
2320 | * The returned facet is guaranteed to be valid. */ | |
2321 | static struct facet * | |
2322 | facet_lookup_valid(struct ofproto *ofproto, const struct flow *flow) | |
2323 | { | |
2324 | struct facet *facet = facet_find(ofproto, flow); | |
2325 | ||
2326 | /* The facet we found might not be valid, since we could be in need of | |
2327 | * revalidation. If it is not valid, don't return it. */ | |
2328 | if (facet | |
2329 | && ofproto->need_revalidate | |
2330 | && !facet_revalidate(ofproto, facet)) { | |
2331 | COVERAGE_INC(ofproto_invalidated); | |
2332 | return NULL; | |
064af421 | 2333 | } |
064af421 | 2334 | |
bcf84111 BP |
2335 | return facet; |
2336 | } | |
0193b2af | 2337 | |
bcf84111 BP |
2338 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': |
2339 | * | |
2340 | * - If the rule found is different from 'facet''s current rule, moves | |
2341 | * 'facet' to the new rule and recompiles its actions. | |
2342 | * | |
2343 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2344 | * where it is and recompiles its actions anyway. | |
2345 | * | |
2346 | * - If there is none, destroys 'facet'. | |
2347 | * | |
d530fcd2 | 2348 | * Returns true if 'facet' still exists, false if it has been destroyed. */ |
bcf84111 BP |
2349 | static bool |
2350 | facet_revalidate(struct ofproto *ofproto, struct facet *facet) | |
2351 | { | |
d530fcd2 BP |
2352 | struct rule *new_rule; |
2353 | struct odp_actions a; | |
2354 | size_t actions_len; | |
2355 | uint16_t new_nf_output_iface; | |
2356 | bool actions_changed; | |
bcf84111 BP |
2357 | |
2358 | COVERAGE_INC(facet_revalidate); | |
d530fcd2 BP |
2359 | |
2360 | /* Determine the new rule. */ | |
2361 | new_rule = rule_lookup(ofproto, &facet->flow); | |
2362 | if (!new_rule) { | |
2363 | /* No new rule, so delete the facet. */ | |
bcf84111 BP |
2364 | facet_remove(ofproto, facet); |
2365 | return false; | |
2366 | } | |
2367 | ||
d530fcd2 BP |
2368 | /* Calculate new ODP actions. |
2369 | * | |
2370 | * We are very cautious about actually modifying 'facet' state at this | |
2371 | * point, because we might need to, e.g., emit a NetFlow expiration and, if | |
2372 | * so, we need to have the old state around to properly compose it. */ | |
2373 | xlate_actions(new_rule->actions, new_rule->n_actions, &facet->flow, | |
2374 | ofproto, NULL, &a, &facet->tags, &facet->may_install, | |
2375 | &new_nf_output_iface); | |
2376 | actions_len = a.n_actions * sizeof *a.actions; | |
2377 | actions_changed = (facet->n_actions != a.n_actions | |
2378 | || memcmp(facet->actions, a.actions, actions_len)); | |
2379 | ||
2380 | /* If the ODP actions changed or the installability changed, then we need | |
2381 | * to talk to the datapath. */ | |
2382 | if (actions_changed || facet->may_install != facet->installed) { | |
2383 | if (facet->may_install) { | |
2384 | struct odp_flow_put put; | |
2385 | ||
2386 | memset(&put.flow.stats, 0, sizeof put.flow.stats); | |
2387 | odp_flow_key_from_flow(&put.flow.key, &facet->flow); | |
2388 | put.flow.actions = a.actions; | |
2389 | put.flow.n_actions = a.n_actions; | |
2390 | put.flow.flags = 0; | |
2391 | put.flags = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS; | |
2392 | dpif_flow_put(ofproto->dpif, &put); | |
2393 | ||
2394 | facet_update_stats(ofproto, facet, &put.flow.stats); | |
2395 | } else { | |
2396 | facet_uninstall(ofproto, facet); | |
2397 | } | |
2398 | ||
2399 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2400 | * 'facet' into 'rule'. */ | |
2401 | facet_flush_stats(ofproto, facet); | |
2402 | } | |
2403 | ||
2404 | /* Update 'facet' now that we've taken care of all the old state. */ | |
2405 | facet->nf_flow.output_iface = new_nf_output_iface; | |
2406 | if (actions_changed) { | |
2407 | free(facet->actions); | |
2408 | facet->n_actions = a.n_actions; | |
2409 | facet->actions = xmemdup(a.actions, actions_len); | |
2410 | } | |
2411 | if (facet->rule != new_rule) { | |
bcf84111 BP |
2412 | COVERAGE_INC(facet_changed_rule); |
2413 | list_remove(&facet->list_node); | |
d530fcd2 BP |
2414 | list_push_back(&new_rule->facets, &facet->list_node); |
2415 | facet->rule = new_rule; | |
2416 | facet->used = new_rule->created; | |
0c0afbec | 2417 | } |
bcf84111 | 2418 | |
bcf84111 | 2419 | return true; |
064af421 BP |
2420 | } |
2421 | \f | |
2422 | static void | |
2423 | queue_tx(struct ofpbuf *msg, const struct ofconn *ofconn, | |
2424 | struct rconn_packet_counter *counter) | |
2425 | { | |
2426 | update_openflow_length(msg); | |
2427 | if (rconn_send(ofconn->rconn, msg, counter)) { | |
2428 | ofpbuf_delete(msg); | |
2429 | } | |
2430 | } | |
2431 | ||
064af421 BP |
2432 | static void |
2433 | send_error_oh(const struct ofconn *ofconn, const struct ofp_header *oh, | |
2434 | int error) | |
2435 | { | |
26c112c2 BP |
2436 | struct ofpbuf *buf = make_ofp_error_msg(error, oh); |
2437 | if (buf) { | |
2438 | COVERAGE_INC(ofproto_error); | |
2439 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2440 | } | |
064af421 BP |
2441 | } |
2442 | ||
2443 | static void | |
2444 | hton_ofp_phy_port(struct ofp_phy_port *opp) | |
2445 | { | |
2446 | opp->port_no = htons(opp->port_no); | |
2447 | opp->config = htonl(opp->config); | |
2448 | opp->state = htonl(opp->state); | |
2449 | opp->curr = htonl(opp->curr); | |
2450 | opp->advertised = htonl(opp->advertised); | |
2451 | opp->supported = htonl(opp->supported); | |
2452 | opp->peer = htonl(opp->peer); | |
2453 | } | |
2454 | ||
2455 | static int | |
2456 | handle_echo_request(struct ofconn *ofconn, struct ofp_header *oh) | |
2457 | { | |
2458 | struct ofp_header *rq = oh; | |
2459 | queue_tx(make_echo_reply(rq), ofconn, ofconn->reply_counter); | |
2460 | return 0; | |
2461 | } | |
2462 | ||
2463 | static int | |
3269c562 | 2464 | handle_features_request(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 BP |
2465 | { |
2466 | struct ofp_switch_features *osf; | |
2467 | struct ofpbuf *buf; | |
064af421 BP |
2468 | struct ofport *port; |
2469 | ||
2470 | osf = make_openflow_xid(sizeof *osf, OFPT_FEATURES_REPLY, oh->xid, &buf); | |
3269c562 | 2471 | osf->datapath_id = htonll(ofconn->ofproto->datapath_id); |
064af421 BP |
2472 | osf->n_buffers = htonl(pktbuf_capacity()); |
2473 | osf->n_tables = 2; | |
2474 | osf->capabilities = htonl(OFPC_FLOW_STATS | OFPC_TABLE_STATS | | |
0254ae23 | 2475 | OFPC_PORT_STATS | OFPC_ARP_MATCH_IP); |
064af421 BP |
2476 | osf->actions = htonl((1u << OFPAT_OUTPUT) | |
2477 | (1u << OFPAT_SET_VLAN_VID) | | |
2478 | (1u << OFPAT_SET_VLAN_PCP) | | |
2479 | (1u << OFPAT_STRIP_VLAN) | | |
2480 | (1u << OFPAT_SET_DL_SRC) | | |
2481 | (1u << OFPAT_SET_DL_DST) | | |
2482 | (1u << OFPAT_SET_NW_SRC) | | |
2483 | (1u << OFPAT_SET_NW_DST) | | |
959a2ecd | 2484 | (1u << OFPAT_SET_NW_TOS) | |
064af421 | 2485 | (1u << OFPAT_SET_TP_SRC) | |
c1c9c9c4 BP |
2486 | (1u << OFPAT_SET_TP_DST) | |
2487 | (1u << OFPAT_ENQUEUE)); | |
064af421 | 2488 | |
3269c562 | 2489 | HMAP_FOR_EACH (port, hmap_node, &ofconn->ofproto->ports) { |
064af421 BP |
2490 | hton_ofp_phy_port(ofpbuf_put(buf, &port->opp, sizeof port->opp)); |
2491 | } | |
2492 | ||
2493 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2494 | return 0; | |
2495 | } | |
2496 | ||
2497 | static int | |
3269c562 | 2498 | handle_get_config_request(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 BP |
2499 | { |
2500 | struct ofpbuf *buf; | |
2501 | struct ofp_switch_config *osc; | |
2502 | uint16_t flags; | |
2503 | bool drop_frags; | |
2504 | ||
2505 | /* Figure out flags. */ | |
3269c562 | 2506 | dpif_get_drop_frags(ofconn->ofproto->dpif, &drop_frags); |
064af421 | 2507 | flags = drop_frags ? OFPC_FRAG_DROP : OFPC_FRAG_NORMAL; |
064af421 BP |
2508 | |
2509 | /* Send reply. */ | |
2510 | osc = make_openflow_xid(sizeof *osc, OFPT_GET_CONFIG_REPLY, oh->xid, &buf); | |
2511 | osc->flags = htons(flags); | |
2512 | osc->miss_send_len = htons(ofconn->miss_send_len); | |
2513 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
2514 | ||
2515 | return 0; | |
2516 | } | |
2517 | ||
2518 | static int | |
3269c562 | 2519 | handle_set_config(struct ofconn *ofconn, struct ofp_switch_config *osc) |
064af421 BP |
2520 | { |
2521 | uint16_t flags; | |
2522 | int error; | |
2523 | ||
2524 | error = check_ofp_message(&osc->header, OFPT_SET_CONFIG, sizeof *osc); | |
2525 | if (error) { | |
2526 | return error; | |
2527 | } | |
2528 | flags = ntohs(osc->flags); | |
2529 | ||
5899143f | 2530 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role != NX_ROLE_SLAVE) { |
064af421 BP |
2531 | switch (flags & OFPC_FRAG_MASK) { |
2532 | case OFPC_FRAG_NORMAL: | |
3269c562 | 2533 | dpif_set_drop_frags(ofconn->ofproto->dpif, false); |
064af421 BP |
2534 | break; |
2535 | case OFPC_FRAG_DROP: | |
3269c562 | 2536 | dpif_set_drop_frags(ofconn->ofproto->dpif, true); |
064af421 BP |
2537 | break; |
2538 | default: | |
2539 | VLOG_WARN_RL(&rl, "requested bad fragment mode (flags=%"PRIx16")", | |
2540 | osc->flags); | |
2541 | break; | |
2542 | } | |
2543 | } | |
2544 | ||
064af421 BP |
2545 | ofconn->miss_send_len = ntohs(osc->miss_send_len); |
2546 | ||
2547 | return 0; | |
2548 | } | |
2549 | ||
064af421 | 2550 | static void |
aae51f53 | 2551 | add_controller_action(struct odp_actions *actions, uint16_t max_len) |
064af421 BP |
2552 | { |
2553 | union odp_action *a = odp_actions_add(actions, ODPAT_CONTROLLER); | |
aae51f53 | 2554 | a->controller.arg = max_len; |
064af421 BP |
2555 | } |
2556 | ||
2557 | struct action_xlate_ctx { | |
2558 | /* Input. */ | |
ae412e7d | 2559 | struct flow flow; /* Flow to which these actions correspond. */ |
064af421 BP |
2560 | int recurse; /* Recursion level, via xlate_table_action. */ |
2561 | struct ofproto *ofproto; | |
2562 | const struct ofpbuf *packet; /* The packet corresponding to 'flow', or a | |
2563 | * null pointer if we are revalidating | |
2564 | * without a packet to refer to. */ | |
2565 | ||
2566 | /* Output. */ | |
2567 | struct odp_actions *out; /* Datapath actions. */ | |
1eb0942d | 2568 | tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ |
d6fbec6d | 2569 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
064af421 | 2570 | * be reassessed for every packet. */ |
6a07af36 | 2571 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
064af421 BP |
2572 | }; |
2573 | ||
5f8bfd69 BP |
2574 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a |
2575 | * flow translation. */ | |
2576 | #define MAX_RESUBMIT_RECURSION 8 | |
2577 | ||
064af421 BP |
2578 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, |
2579 | struct action_xlate_ctx *ctx); | |
2580 | ||
2581 | static void | |
2582 | add_output_action(struct action_xlate_ctx *ctx, uint16_t port) | |
2583 | { | |
ca0f572c | 2584 | const struct ofport *ofport = get_port(ctx->ofproto, port); |
6cfaf517 BP |
2585 | |
2586 | if (ofport) { | |
2587 | if (ofport->opp.config & OFPPC_NO_FWD) { | |
2588 | /* Forwarding disabled on port. */ | |
2589 | return; | |
2590 | } | |
2591 | } else { | |
2592 | /* | |
2593 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2594 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2595 | * and we're pre-populating the flow table. | |
2596 | */ | |
064af421 | 2597 | } |
6cfaf517 BP |
2598 | |
2599 | odp_actions_add(ctx->out, ODPAT_OUTPUT)->output.port = port; | |
6a07af36 | 2600 | ctx->nf_output_iface = port; |
064af421 BP |
2601 | } |
2602 | ||
2603 | static struct rule * | |
bcf84111 | 2604 | rule_lookup(struct ofproto *ofproto, const struct flow *flow) |
064af421 | 2605 | { |
3c4486a5 | 2606 | return rule_from_cls_rule(classifier_lookup(&ofproto->cls, flow)); |
064af421 BP |
2607 | } |
2608 | ||
2609 | static void | |
2610 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2611 | { | |
5f8bfd69 | 2612 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { |
2c5d1389 | 2613 | uint16_t old_in_port; |
064af421 | 2614 | struct rule *rule; |
064af421 | 2615 | |
2c5d1389 BP |
2616 | /* Look up a flow with 'in_port' as the input port. Then restore the |
2617 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2618 | * have surprising behavior). */ | |
2619 | old_in_port = ctx->flow.in_port; | |
e18fe8a2 | 2620 | ctx->flow.in_port = in_port; |
bcf84111 | 2621 | rule = rule_lookup(ctx->ofproto, &ctx->flow); |
2c5d1389 BP |
2622 | ctx->flow.in_port = old_in_port; |
2623 | ||
064af421 | 2624 | if (rule) { |
064af421 BP |
2625 | ctx->recurse++; |
2626 | do_xlate_actions(rule->actions, rule->n_actions, ctx); | |
2627 | ctx->recurse--; | |
2628 | } | |
5f8bfd69 BP |
2629 | } else { |
2630 | struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
2631 | ||
2632 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2633 | MAX_RESUBMIT_RECURSION); | |
064af421 BP |
2634 | } |
2635 | } | |
2636 | ||
f1588b1f BP |
2637 | static void |
2638 | flood_packets(struct ofproto *ofproto, uint16_t odp_in_port, uint32_t mask, | |
2639 | uint16_t *nf_output_iface, struct odp_actions *actions) | |
2640 | { | |
2641 | struct ofport *ofport; | |
2642 | ||
2643 | HMAP_FOR_EACH (ofport, hmap_node, &ofproto->ports) { | |
2644 | uint16_t odp_port = ofport->odp_port; | |
2645 | if (odp_port != odp_in_port && !(ofport->opp.config & mask)) { | |
2646 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = odp_port; | |
2647 | } | |
2648 | } | |
2649 | *nf_output_iface = NF_OUT_FLOOD; | |
2650 | } | |
2651 | ||
064af421 | 2652 | static void |
aae51f53 BP |
2653 | xlate_output_action__(struct action_xlate_ctx *ctx, |
2654 | uint16_t port, uint16_t max_len) | |
064af421 BP |
2655 | { |
2656 | uint16_t odp_port; | |
6a07af36 JG |
2657 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; |
2658 | ||
2659 | ctx->nf_output_iface = NF_OUT_DROP; | |
064af421 | 2660 | |
aae51f53 | 2661 | switch (port) { |
064af421 | 2662 | case OFPP_IN_PORT: |
e18fe8a2 | 2663 | add_output_action(ctx, ctx->flow.in_port); |
064af421 BP |
2664 | break; |
2665 | case OFPP_TABLE: | |
e18fe8a2 | 2666 | xlate_table_action(ctx, ctx->flow.in_port); |
064af421 BP |
2667 | break; |
2668 | case OFPP_NORMAL: | |
e18fe8a2 | 2669 | if (!ctx->ofproto->ofhooks->normal_cb(&ctx->flow, ctx->packet, |
1eb0942d | 2670 | ctx->out, &ctx->tags, |
6a07af36 | 2671 | &ctx->nf_output_iface, |
064af421 BP |
2672 | ctx->ofproto->aux)) { |
2673 | COVERAGE_INC(ofproto_uninstallable); | |
d6fbec6d | 2674 | ctx->may_set_up_flow = false; |
064af421 BP |
2675 | } |
2676 | break; | |
2677 | case OFPP_FLOOD: | |
f1588b1f BP |
2678 | flood_packets(ctx->ofproto, ctx->flow.in_port, OFPPC_NO_FLOOD, |
2679 | &ctx->nf_output_iface, ctx->out); | |
9628cd42 | 2680 | break; |
064af421 | 2681 | case OFPP_ALL: |
f1588b1f BP |
2682 | flood_packets(ctx->ofproto, ctx->flow.in_port, 0, |
2683 | &ctx->nf_output_iface, ctx->out); | |
064af421 BP |
2684 | break; |
2685 | case OFPP_CONTROLLER: | |
aae51f53 | 2686 | add_controller_action(ctx->out, max_len); |
064af421 BP |
2687 | break; |
2688 | case OFPP_LOCAL: | |
2689 | add_output_action(ctx, ODPP_LOCAL); | |
2690 | break; | |
2691 | default: | |
aae51f53 | 2692 | odp_port = ofp_port_to_odp_port(port); |
e18fe8a2 | 2693 | if (odp_port != ctx->flow.in_port) { |
064af421 BP |
2694 | add_output_action(ctx, odp_port); |
2695 | } | |
2696 | break; | |
2697 | } | |
6a07af36 JG |
2698 | |
2699 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2700 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2701 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2702 | ctx->nf_output_iface = prev_nf_output_iface; | |
2703 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2704 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2705 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2706 | } | |
064af421 BP |
2707 | } |
2708 | ||
aae51f53 BP |
2709 | static void |
2710 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2711 | const struct ofp_action_output *oao) | |
2712 | { | |
2713 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2714 | } | |
2715 | ||
c1c9c9c4 BP |
2716 | /* If the final ODP action in 'ctx' is "pop priority", drop it, as an |
2717 | * optimization, because we're going to add another action that sets the | |
2718 | * priority immediately after, or because there are no actions following the | |
2719 | * pop. */ | |
2720 | static void | |
2721 | remove_pop_action(struct action_xlate_ctx *ctx) | |
2722 | { | |
2723 | size_t n = ctx->out->n_actions; | |
2724 | if (n > 0 && ctx->out->actions[n - 1].type == ODPAT_POP_PRIORITY) { | |
2725 | ctx->out->n_actions--; | |
2726 | } | |
2727 | } | |
2728 | ||
2729 | static void | |
2730 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2731 | const struct ofp_action_enqueue *oae) | |
2732 | { | |
2733 | uint16_t ofp_port, odp_port; | |
aae51f53 BP |
2734 | uint32_t priority; |
2735 | int error; | |
2736 | ||
2737 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2738 | &priority); | |
2739 | if (error) { | |
2740 | /* Fall back to ordinary output action. */ | |
2741 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
2742 | return; | |
2743 | } | |
c1c9c9c4 BP |
2744 | |
2745 | /* Figure out ODP output port. */ | |
2746 | ofp_port = ntohs(oae->port); | |
2747 | if (ofp_port != OFPP_IN_PORT) { | |
2748 | odp_port = ofp_port_to_odp_port(ofp_port); | |
2749 | } else { | |
2750 | odp_port = ctx->flow.in_port; | |
2751 | } | |
2752 | ||
2753 | /* Add ODP actions. */ | |
2754 | remove_pop_action(ctx); | |
2755 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
aae51f53 | 2756 | = priority; |
c1c9c9c4 BP |
2757 | add_output_action(ctx, odp_port); |
2758 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2759 | ||
2760 | /* Update NetFlow output port. */ | |
2761 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2762 | ctx->nf_output_iface = odp_port; | |
2763 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2764 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2765 | } | |
2766 | } | |
2767 | ||
eedc0097 JP |
2768 | static void |
2769 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
2770 | const struct nx_action_set_queue *nasq) | |
2771 | { | |
2772 | uint32_t priority; | |
2773 | int error; | |
2774 | ||
2775 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
2776 | &priority); | |
2777 | if (error) { | |
2778 | /* Couldn't translate queue to a priority, so ignore. A warning | |
2779 | * has already been logged. */ | |
2780 | return; | |
2781 | } | |
2782 | ||
2783 | remove_pop_action(ctx); | |
2784 | odp_actions_add(ctx->out, ODPAT_SET_PRIORITY)->priority.priority | |
2785 | = priority; | |
2786 | } | |
2787 | ||
350a665f BP |
2788 | static void |
2789 | xlate_set_dl_tci(struct action_xlate_ctx *ctx) | |
2790 | { | |
66642cb4 BP |
2791 | ovs_be16 tci = ctx->flow.vlan_tci; |
2792 | if (!(tci & htons(VLAN_CFI))) { | |
350a665f BP |
2793 | odp_actions_add(ctx->out, ODPAT_STRIP_VLAN); |
2794 | } else { | |
2795 | union odp_action *oa = odp_actions_add(ctx->out, ODPAT_SET_DL_TCI); | |
66642cb4 | 2796 | oa->dl_tci.tci = tci & ~htons(VLAN_CFI); |
350a665f BP |
2797 | } |
2798 | } | |
2799 | ||
b6c9e612 BP |
2800 | static void |
2801 | xlate_reg_move_action(struct action_xlate_ctx *ctx, | |
2802 | const struct nx_action_reg_move *narm) | |
2803 | { | |
66642cb4 | 2804 | ovs_be16 old_tci = ctx->flow.vlan_tci; |
b6c9e612 BP |
2805 | |
2806 | nxm_execute_reg_move(narm, &ctx->flow); | |
2807 | ||
66642cb4 | 2808 | if (ctx->flow.vlan_tci != old_tci) { |
b6c9e612 BP |
2809 | xlate_set_dl_tci(ctx); |
2810 | } | |
2811 | } | |
2812 | ||
064af421 BP |
2813 | static void |
2814 | xlate_nicira_action(struct action_xlate_ctx *ctx, | |
2815 | const struct nx_action_header *nah) | |
2816 | { | |
2817 | const struct nx_action_resubmit *nar; | |
659586ef | 2818 | const struct nx_action_set_tunnel *nast; |
eedc0097 | 2819 | const struct nx_action_set_queue *nasq; |
659586ef | 2820 | union odp_action *oa; |
064af421 BP |
2821 | int subtype = ntohs(nah->subtype); |
2822 | ||
2823 | assert(nah->vendor == htonl(NX_VENDOR_ID)); | |
2824 | switch (subtype) { | |
2825 | case NXAST_RESUBMIT: | |
2826 | nar = (const struct nx_action_resubmit *) nah; | |
2827 | xlate_table_action(ctx, ofp_port_to_odp_port(ntohs(nar->in_port))); | |
2828 | break; | |
2829 | ||
659586ef JG |
2830 | case NXAST_SET_TUNNEL: |
2831 | nast = (const struct nx_action_set_tunnel *) nah; | |
2832 | oa = odp_actions_add(ctx->out, ODPAT_SET_TUNNEL); | |
2833 | ctx->flow.tun_id = oa->tunnel.tun_id = nast->tun_id; | |
2834 | break; | |
2835 | ||
401eeb92 BP |
2836 | case NXAST_DROP_SPOOFED_ARP: |
2837 | if (ctx->flow.dl_type == htons(ETH_TYPE_ARP)) { | |
2838 | odp_actions_add(ctx->out, ODPAT_DROP_SPOOFED_ARP); | |
2839 | } | |
2840 | break; | |
2841 | ||
eedc0097 JP |
2842 | case NXAST_SET_QUEUE: |
2843 | nasq = (const struct nx_action_set_queue *) nah; | |
2844 | xlate_set_queue_action(ctx, nasq); | |
2845 | break; | |
2846 | ||
2847 | case NXAST_POP_QUEUE: | |
2848 | odp_actions_add(ctx->out, ODPAT_POP_PRIORITY); | |
2849 | break; | |
2850 | ||
b6c9e612 BP |
2851 | case NXAST_REG_MOVE: |
2852 | xlate_reg_move_action(ctx, (const struct nx_action_reg_move *) nah); | |
2853 | break; | |
2854 | ||
2855 | case NXAST_REG_LOAD: | |
2856 | nxm_execute_reg_load((const struct nx_action_reg_load *) nah, | |
2857 | &ctx->flow); | |
96fc46e8 BP |
2858 | |
2859 | case NXAST_NOTE: | |
2860 | /* Nothing to do. */ | |
b6c9e612 BP |
2861 | break; |
2862 | ||
999f0d45 | 2863 | /* If you add a new action here that modifies flow data, don't forget to |
c1c9c9c4 | 2864 | * update the flow key in ctx->flow at the same time. */ |
999f0d45 | 2865 | |
064af421 BP |
2866 | default: |
2867 | VLOG_DBG_RL(&rl, "unknown Nicira action type %"PRIu16, subtype); | |
2868 | break; | |
2869 | } | |
2870 | } | |
2871 | ||
2872 | static void | |
2873 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2874 | struct action_xlate_ctx *ctx) | |
2875 | { | |
2876 | struct actions_iterator iter; | |
2877 | const union ofp_action *ia; | |
2878 | const struct ofport *port; | |
2879 | ||
ca0f572c | 2880 | port = get_port(ctx->ofproto, ctx->flow.in_port); |
064af421 | 2881 | if (port && port->opp.config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && |
ba186119 | 2882 | port->opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) |
064af421 BP |
2883 | ? OFPPC_NO_RECV_STP : OFPPC_NO_RECV)) { |
2884 | /* Drop this flow. */ | |
2885 | return; | |
2886 | } | |
2887 | ||
2888 | for (ia = actions_first(&iter, in, n_in); ia; ia = actions_next(&iter)) { | |
2889 | uint16_t type = ntohs(ia->type); | |
2890 | union odp_action *oa; | |
2891 | ||
2892 | switch (type) { | |
2893 | case OFPAT_OUTPUT: | |
2894 | xlate_output_action(ctx, &ia->output); | |
2895 | break; | |
2896 | ||
2897 | case OFPAT_SET_VLAN_VID: | |
66642cb4 BP |
2898 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
2899 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
350a665f | 2900 | xlate_set_dl_tci(ctx); |
064af421 BP |
2901 | break; |
2902 | ||
2903 | case OFPAT_SET_VLAN_PCP: | |
66642cb4 BP |
2904 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
2905 | ctx->flow.vlan_tci |= htons( | |
2906 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
350a665f | 2907 | xlate_set_dl_tci(ctx); |
064af421 BP |
2908 | break; |
2909 | ||
2910 | case OFPAT_STRIP_VLAN: | |
66642cb4 | 2911 | ctx->flow.vlan_tci = htons(0); |
350a665f | 2912 | xlate_set_dl_tci(ctx); |
064af421 BP |
2913 | break; |
2914 | ||
2915 | case OFPAT_SET_DL_SRC: | |
2916 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_SRC); | |
2917 | memcpy(oa->dl_addr.dl_addr, | |
2918 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2919 | memcpy(ctx->flow.dl_src, |
2920 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2921 | break; |
2922 | ||
2923 | case OFPAT_SET_DL_DST: | |
2924 | oa = odp_actions_add(ctx->out, ODPAT_SET_DL_DST); | |
2925 | memcpy(oa->dl_addr.dl_addr, | |
2926 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
999f0d45 BP |
2927 | memcpy(ctx->flow.dl_dst, |
2928 | ((struct ofp_action_dl_addr *) ia)->dl_addr, ETH_ADDR_LEN); | |
064af421 BP |
2929 | break; |
2930 | ||
2931 | case OFPAT_SET_NW_SRC: | |
2932 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_SRC); | |
999f0d45 | 2933 | ctx->flow.nw_src = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
064af421 BP |
2934 | break; |
2935 | ||
2d70a31a JP |
2936 | case OFPAT_SET_NW_DST: |
2937 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_DST); | |
999f0d45 | 2938 | ctx->flow.nw_dst = oa->nw_addr.nw_addr = ia->nw_addr.nw_addr; |
2d38e234 | 2939 | break; |
959a2ecd JP |
2940 | |
2941 | case OFPAT_SET_NW_TOS: | |
2942 | oa = odp_actions_add(ctx->out, ODPAT_SET_NW_TOS); | |
999f0d45 | 2943 | ctx->flow.nw_tos = oa->nw_tos.nw_tos = ia->nw_tos.nw_tos; |
2d70a31a JP |
2944 | break; |
2945 | ||
064af421 BP |
2946 | case OFPAT_SET_TP_SRC: |
2947 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_SRC); | |
999f0d45 | 2948 | ctx->flow.tp_src = oa->tp_port.tp_port = ia->tp_port.tp_port; |
064af421 BP |
2949 | break; |
2950 | ||
2d70a31a JP |
2951 | case OFPAT_SET_TP_DST: |
2952 | oa = odp_actions_add(ctx->out, ODPAT_SET_TP_DST); | |
999f0d45 | 2953 | ctx->flow.tp_dst = oa->tp_port.tp_port = ia->tp_port.tp_port; |
2d70a31a JP |
2954 | break; |
2955 | ||
064af421 BP |
2956 | case OFPAT_VENDOR: |
2957 | xlate_nicira_action(ctx, (const struct nx_action_header *) ia); | |
2958 | break; | |
2959 | ||
c1c9c9c4 BP |
2960 | case OFPAT_ENQUEUE: |
2961 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
2962 | break; | |
2963 | ||
064af421 BP |
2964 | default: |
2965 | VLOG_DBG_RL(&rl, "unknown action type %"PRIu16, type); | |
2966 | break; | |
2967 | } | |
2968 | } | |
2969 | } | |
2970 | ||
2971 | static int | |
2972 | xlate_actions(const union ofp_action *in, size_t n_in, | |
ae412e7d | 2973 | const struct flow *flow, struct ofproto *ofproto, |
064af421 | 2974 | const struct ofpbuf *packet, |
6a07af36 JG |
2975 | struct odp_actions *out, tag_type *tags, bool *may_set_up_flow, |
2976 | uint16_t *nf_output_iface) | |
064af421 | 2977 | { |
064af421 | 2978 | struct action_xlate_ctx ctx; |
1eb0942d | 2979 | |
064af421 BP |
2980 | COVERAGE_INC(ofproto_ofp2odp); |
2981 | odp_actions_init(out); | |
e18fe8a2 | 2982 | ctx.flow = *flow; |
064af421 BP |
2983 | ctx.recurse = 0; |
2984 | ctx.ofproto = ofproto; | |
2985 | ctx.packet = packet; | |
2986 | ctx.out = out; | |
1eb0942d | 2987 | ctx.tags = 0; |
d6fbec6d | 2988 | ctx.may_set_up_flow = true; |
6a07af36 | 2989 | ctx.nf_output_iface = NF_OUT_DROP; |
064af421 | 2990 | do_xlate_actions(in, n_in, &ctx); |
c1c9c9c4 | 2991 | remove_pop_action(&ctx); |
0ad9b732 | 2992 | |
d6fbec6d | 2993 | /* Check with in-band control to see if we're allowed to set up this |
0ad9b732 JP |
2994 | * flow. */ |
2995 | if (!in_band_rule_check(ofproto->in_band, flow, out)) { | |
d6fbec6d | 2996 | ctx.may_set_up_flow = false; |
0ad9b732 JP |
2997 | } |
2998 | ||
1eb0942d BP |
2999 | if (tags) { |
3000 | *tags = ctx.tags; | |
3001 | } | |
d6fbec6d BP |
3002 | if (may_set_up_flow) { |
3003 | *may_set_up_flow = ctx.may_set_up_flow; | |
064af421 | 3004 | } |
6a07af36 JG |
3005 | if (nf_output_iface) { |
3006 | *nf_output_iface = ctx.nf_output_iface; | |
064af421 BP |
3007 | } |
3008 | if (odp_actions_overflow(out)) { | |
9026bf34 | 3009 | COVERAGE_INC(odp_overflow); |
064af421 BP |
3010 | odp_actions_init(out); |
3011 | return ofp_mkerr(OFPET_BAD_ACTION, OFPBAC_TOO_MANY); | |
3012 | } | |
3013 | return 0; | |
3014 | } | |
3015 | ||
9deba63b BP |
3016 | /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow |
3017 | * error message code (composed with ofp_mkerr()) for the caller to propagate | |
3018 | * upward. Otherwise, returns 0. | |
3019 | * | |
2228b50d | 3020 | * The log message mentions 'msg_type'. */ |
9deba63b | 3021 | static int |
2228b50d | 3022 | reject_slave_controller(struct ofconn *ofconn, const const char *msg_type) |
9deba63b | 3023 | { |
5899143f | 3024 | if (ofconn->type == OFCONN_PRIMARY && ofconn->role == NX_ROLE_SLAVE) { |
9deba63b | 3025 | static struct vlog_rate_limit perm_rl = VLOG_RATE_LIMIT_INIT(1, 5); |
9deba63b | 3026 | VLOG_WARN_RL(&perm_rl, "rejecting %s message from slave controller", |
2228b50d | 3027 | msg_type); |
9deba63b BP |
3028 | |
3029 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
3030 | } else { | |
3031 | return 0; | |
3032 | } | |
3033 | } | |
3034 | ||
064af421 | 3035 | static int |
3269c562 | 3036 | handle_packet_out(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 | 3037 | { |
3269c562 | 3038 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3039 | struct ofp_packet_out *opo; |
3040 | struct ofpbuf payload, *buffer; | |
ac51afaf BP |
3041 | union ofp_action *ofp_actions; |
3042 | struct odp_actions odp_actions; | |
3043 | struct ofpbuf request; | |
ae412e7d | 3044 | struct flow flow; |
ac51afaf | 3045 | size_t n_ofp_actions; |
064af421 | 3046 | uint16_t in_port; |
064af421 BP |
3047 | int error; |
3048 | ||
ac51afaf BP |
3049 | COVERAGE_INC(ofproto_packet_out); |
3050 | ||
2228b50d | 3051 | error = reject_slave_controller(ofconn, "OFPT_PACKET_OUT"); |
9deba63b BP |
3052 | if (error) { |
3053 | return error; | |
3054 | } | |
3055 | ||
ac51afaf BP |
3056 | /* Get ofp_packet_out. */ |
3057 | request.data = oh; | |
3058 | request.size = ntohs(oh->length); | |
3059 | opo = ofpbuf_try_pull(&request, offsetof(struct ofp_packet_out, actions)); | |
3060 | if (!opo) { | |
3061 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3062 | } | |
3063 | ||
3064 | /* Get actions. */ | |
3065 | error = ofputil_pull_actions(&request, ntohs(opo->actions_len), | |
3066 | &ofp_actions, &n_ofp_actions); | |
064af421 BP |
3067 | if (error) { |
3068 | return error; | |
3069 | } | |
064af421 | 3070 | |
ac51afaf | 3071 | /* Get payload. */ |
064af421 BP |
3072 | if (opo->buffer_id != htonl(UINT32_MAX)) { |
3073 | error = pktbuf_retrieve(ofconn->pktbuf, ntohl(opo->buffer_id), | |
3074 | &buffer, &in_port); | |
7778bd15 | 3075 | if (error || !buffer) { |
064af421 BP |
3076 | return error; |
3077 | } | |
3078 | payload = *buffer; | |
3079 | } else { | |
ac51afaf | 3080 | payload = request; |
064af421 BP |
3081 | buffer = NULL; |
3082 | } | |
3083 | ||
ac51afaf BP |
3084 | /* Extract flow, check actions. */ |
3085 | flow_extract(&payload, 0, ofp_port_to_odp_port(ntohs(opo->in_port)), | |
3086 | &flow); | |
f1defbf9 | 3087 | error = validate_actions(ofp_actions, n_ofp_actions, &flow, p->max_ports); |
ac51afaf BP |
3088 | if (error) { |
3089 | goto exit; | |
3090 | } | |
3091 | ||
3092 | /* Send. */ | |
3093 | error = xlate_actions(ofp_actions, n_ofp_actions, &flow, p, &payload, | |
3094 | &odp_actions, NULL, NULL, NULL); | |
700712e2 | 3095 | if (!error) { |
ac51afaf BP |
3096 | dpif_execute(p->dpif, odp_actions.actions, odp_actions.n_actions, |
3097 | &payload); | |
064af421 | 3098 | } |
064af421 | 3099 | |
ac51afaf BP |
3100 | exit: |
3101 | ofpbuf_delete(buffer); | |
3102 | return 0; | |
064af421 BP |
3103 | } |
3104 | ||
3105 | static void | |
3106 | update_port_config(struct ofproto *p, struct ofport *port, | |
3107 | uint32_t config, uint32_t mask) | |
3108 | { | |
3109 | mask &= config ^ port->opp.config; | |
3110 | if (mask & OFPPC_PORT_DOWN) { | |
3111 | if (config & OFPPC_PORT_DOWN) { | |
3112 | netdev_turn_flags_off(port->netdev, NETDEV_UP, true); | |
3113 | } else { | |
3114 | netdev_turn_flags_on(port->netdev, NETDEV_UP, true); | |
3115 | } | |
3116 | } | |
f1588b1f BP |
3117 | #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \ |
3118 | OFPPC_NO_FWD | OFPPC_NO_FLOOD) | |
064af421 BP |
3119 | if (mask & REVALIDATE_BITS) { |
3120 | COVERAGE_INC(ofproto_costly_flags); | |
3121 | port->opp.config ^= mask & REVALIDATE_BITS; | |
3122 | p->need_revalidate = true; | |
3123 | } | |
3124 | #undef REVALIDATE_BITS | |
064af421 BP |
3125 | if (mask & OFPPC_NO_PACKET_IN) { |
3126 | port->opp.config ^= OFPPC_NO_PACKET_IN; | |
3127 | } | |
3128 | } | |
3129 | ||
3130 | static int | |
3269c562 | 3131 | handle_port_mod(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 | 3132 | { |
3269c562 | 3133 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3134 | const struct ofp_port_mod *opm; |
3135 | struct ofport *port; | |
3136 | int error; | |
3137 | ||
2228b50d | 3138 | error = reject_slave_controller(ofconn, "OFPT_PORT_MOD"); |
9deba63b BP |
3139 | if (error) { |
3140 | return error; | |
3141 | } | |
064af421 BP |
3142 | error = check_ofp_message(oh, OFPT_PORT_MOD, sizeof *opm); |
3143 | if (error) { | |
3144 | return error; | |
3145 | } | |
3146 | opm = (struct ofp_port_mod *) oh; | |
3147 | ||
ca0f572c | 3148 | port = get_port(p, ofp_port_to_odp_port(ntohs(opm->port_no))); |
064af421 BP |
3149 | if (!port) { |
3150 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_PORT); | |
3151 | } else if (memcmp(port->opp.hw_addr, opm->hw_addr, OFP_ETH_ALEN)) { | |
3152 | return ofp_mkerr(OFPET_PORT_MOD_FAILED, OFPPMFC_BAD_HW_ADDR); | |
3153 | } else { | |
3154 | update_port_config(p, port, ntohl(opm->config), ntohl(opm->mask)); | |
3155 | if (opm->advertise) { | |
3156 | netdev_set_advertisements(port->netdev, ntohl(opm->advertise)); | |
3157 | } | |
3158 | } | |
3159 | return 0; | |
3160 | } | |
3161 | ||
3162 | static struct ofpbuf * | |
06a5e131 | 3163 | make_ofp_stats_reply(ovs_be32 xid, ovs_be16 type, size_t body_len) |
064af421 BP |
3164 | { |
3165 | struct ofp_stats_reply *osr; | |
3166 | struct ofpbuf *msg; | |
3167 | ||
3168 | msg = ofpbuf_new(MIN(sizeof *osr + body_len, UINT16_MAX)); | |
3169 | osr = put_openflow_xid(sizeof *osr, OFPT_STATS_REPLY, xid, msg); | |
3170 | osr->type = type; | |
3171 | osr->flags = htons(0); | |
3172 | return msg; | |
3173 | } | |
3174 | ||
3175 | static struct ofpbuf * | |
06a5e131 | 3176 | start_ofp_stats_reply(const struct ofp_stats_request *request, size_t body_len) |
064af421 | 3177 | { |
06a5e131 | 3178 | return make_ofp_stats_reply(request->header.xid, request->type, body_len); |
064af421 BP |
3179 | } |
3180 | ||
3181 | static void * | |
06a5e131 BP |
3182 | append_ofp_stats_reply(size_t nbytes, struct ofconn *ofconn, |
3183 | struct ofpbuf **msgp) | |
064af421 BP |
3184 | { |
3185 | struct ofpbuf *msg = *msgp; | |
3186 | assert(nbytes <= UINT16_MAX - sizeof(struct ofp_stats_reply)); | |
3187 | if (nbytes + msg->size > UINT16_MAX) { | |
3188 | struct ofp_stats_reply *reply = msg->data; | |
3189 | reply->flags = htons(OFPSF_REPLY_MORE); | |
06a5e131 | 3190 | *msgp = make_ofp_stats_reply(reply->header.xid, reply->type, nbytes); |
064af421 BP |
3191 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3192 | } | |
3193 | return ofpbuf_put_uninit(*msgp, nbytes); | |
3194 | } | |
3195 | ||
09246b99 BP |
3196 | static struct ofpbuf * |
3197 | make_nxstats_reply(ovs_be32 xid, ovs_be32 subtype, size_t body_len) | |
3198 | { | |
3199 | struct nicira_stats_msg *nsm; | |
3200 | struct ofpbuf *msg; | |
3201 | ||
3202 | msg = ofpbuf_new(MIN(sizeof *nsm + body_len, UINT16_MAX)); | |
3203 | nsm = put_openflow_xid(sizeof *nsm, OFPT_STATS_REPLY, xid, msg); | |
3204 | nsm->type = htons(OFPST_VENDOR); | |
3205 | nsm->flags = htons(0); | |
3206 | nsm->vendor = htonl(NX_VENDOR_ID); | |
3207 | nsm->subtype = htonl(subtype); | |
3208 | return msg; | |
3209 | } | |
3210 | ||
3211 | static struct ofpbuf * | |
3212 | start_nxstats_reply(const struct nicira_stats_msg *request, size_t body_len) | |
3213 | { | |
3214 | return make_nxstats_reply(request->header.xid, request->subtype, body_len); | |
3215 | } | |
3216 | ||
3217 | static void | |
3218 | append_nxstats_reply(size_t nbytes, struct ofconn *ofconn, | |
3219 | struct ofpbuf **msgp) | |
3220 | { | |
3221 | struct ofpbuf *msg = *msgp; | |
3222 | assert(nbytes <= UINT16_MAX - sizeof(struct nicira_stats_msg)); | |
3223 | if (nbytes + msg->size > UINT16_MAX) { | |
3224 | struct nicira_stats_msg *reply = msg->data; | |
3225 | reply->flags = htons(OFPSF_REPLY_MORE); | |
3226 | *msgp = make_nxstats_reply(reply->header.xid, reply->subtype, nbytes); | |
3227 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3228 | } | |
3229 | ofpbuf_prealloc_tailroom(*msgp, nbytes); | |
3230 | } | |
3231 | ||
064af421 | 3232 | static int |
3269c562 BP |
3233 | handle_desc_stats_request(struct ofconn *ofconn, |
3234 | struct ofp_stats_request *request) | |
064af421 | 3235 | { |
3269c562 | 3236 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3237 | struct ofp_desc_stats *ods; |
3238 | struct ofpbuf *msg; | |
3239 | ||
06a5e131 BP |
3240 | msg = start_ofp_stats_reply(request, sizeof *ods); |
3241 | ods = append_ofp_stats_reply(sizeof *ods, ofconn, &msg); | |
5a719c38 JP |
3242 | memset(ods, 0, sizeof *ods); |
3243 | ovs_strlcpy(ods->mfr_desc, p->mfr_desc, sizeof ods->mfr_desc); | |
3244 | ovs_strlcpy(ods->hw_desc, p->hw_desc, sizeof ods->hw_desc); | |
3245 | ovs_strlcpy(ods->sw_desc, p->sw_desc, sizeof ods->sw_desc); | |
3246 | ovs_strlcpy(ods->serial_num, p->serial_desc, sizeof ods->serial_num); | |
3247 | ovs_strlcpy(ods->dp_desc, p->dp_desc, sizeof ods->dp_desc); | |
064af421 BP |
3248 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3249 | ||
3250 | return 0; | |
3251 | } | |
3252 | ||
064af421 | 3253 | static int |
3269c562 | 3254 | handle_table_stats_request(struct ofconn *ofconn, |
064af421 BP |
3255 | struct ofp_stats_request *request) |
3256 | { | |
3269c562 | 3257 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3258 | struct ofp_table_stats *ots; |
3259 | struct ofpbuf *msg; | |
064af421 | 3260 | |
06a5e131 | 3261 | msg = start_ofp_stats_reply(request, sizeof *ots * 2); |
064af421 | 3262 | |
064af421 | 3263 | /* Classifier table. */ |
06a5e131 | 3264 | ots = append_ofp_stats_reply(sizeof *ots, ofconn, &msg); |
064af421 | 3265 | memset(ots, 0, sizeof *ots); |
064af421 | 3266 | strcpy(ots->name, "classifier"); |
b70eac89 | 3267 | ots->wildcards = (ofconn->flow_format == NXFF_OPENFLOW10 |
f9bfea14 | 3268 | ? htonl(OFPFW_ALL) : htonl(OVSFW_ALL)); |
ad828225 | 3269 | ots->max_entries = htonl(1024 * 1024); /* An arbitrary big number. */ |
bcf84111 | 3270 | ots->active_count = htonl(classifier_count(&p->cls)); |
064af421 BP |
3271 | ots->lookup_count = htonll(0); /* XXX */ |
3272 | ots->matched_count = htonll(0); /* XXX */ | |
3273 | ||
3274 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3275 | return 0; | |
3276 | } | |
3277 | ||
abaad8cf | 3278 | static void |
ca0f572c | 3279 | append_port_stat(struct ofport *port, struct ofconn *ofconn, |
a4948b95 | 3280 | struct ofpbuf **msgp) |
abaad8cf JP |
3281 | { |
3282 | struct netdev_stats stats; | |
3283 | struct ofp_port_stats *ops; | |
3284 | ||
d295e8e9 JP |
3285 | /* Intentionally ignore return value, since errors will set |
3286 | * 'stats' to all-1s, which is correct for OpenFlow, and | |
abaad8cf JP |
3287 | * netdev_get_stats() will log errors. */ |
3288 | netdev_get_stats(port->netdev, &stats); | |
3289 | ||
06a5e131 | 3290 | ops = append_ofp_stats_reply(sizeof *ops, ofconn, msgp); |
ca0f572c | 3291 | ops->port_no = htons(port->opp.port_no); |
abaad8cf JP |
3292 | memset(ops->pad, 0, sizeof ops->pad); |
3293 | ops->rx_packets = htonll(stats.rx_packets); | |
3294 | ops->tx_packets = htonll(stats.tx_packets); | |
3295 | ops->rx_bytes = htonll(stats.rx_bytes); | |
3296 | ops->tx_bytes = htonll(stats.tx_bytes); | |
3297 | ops->rx_dropped = htonll(stats.rx_dropped); | |
3298 | ops->tx_dropped = htonll(stats.tx_dropped); | |
3299 | ops->rx_errors = htonll(stats.rx_errors); | |
3300 | ops->tx_errors = htonll(stats.tx_errors); | |
3301 | ops->rx_frame_err = htonll(stats.rx_frame_errors); | |
3302 | ops->rx_over_err = htonll(stats.rx_over_errors); | |
3303 | ops->rx_crc_err = htonll(stats.rx_crc_errors); | |
3304 | ops->collisions = htonll(stats.collisions); | |
3305 | } | |
3306 | ||
064af421 | 3307 | static int |
3269c562 | 3308 | handle_port_stats_request(struct ofconn *ofconn, struct ofp_stats_request *osr, |
abaad8cf | 3309 | size_t arg_size) |
064af421 | 3310 | { |
3269c562 | 3311 | struct ofproto *p = ofconn->ofproto; |
abaad8cf | 3312 | struct ofp_port_stats_request *psr; |
064af421 BP |
3313 | struct ofp_port_stats *ops; |
3314 | struct ofpbuf *msg; | |
3315 | struct ofport *port; | |
064af421 | 3316 | |
abaad8cf JP |
3317 | if (arg_size != sizeof *psr) { |
3318 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3319 | } | |
3320 | psr = (struct ofp_port_stats_request *) osr->body; | |
3321 | ||
06a5e131 | 3322 | msg = start_ofp_stats_reply(osr, sizeof *ops * 16); |
abaad8cf | 3323 | if (psr->port_no != htons(OFPP_NONE)) { |
ca0f572c | 3324 | port = get_port(p, ofp_port_to_odp_port(ntohs(psr->port_no))); |
abaad8cf | 3325 | if (port) { |
ca0f572c | 3326 | append_port_stat(port, ofconn, &msg); |
abaad8cf JP |
3327 | } |
3328 | } else { | |
4e8e4213 | 3329 | HMAP_FOR_EACH (port, hmap_node, &p->ports) { |
ca0f572c | 3330 | append_port_stat(port, ofconn, &msg); |
abaad8cf | 3331 | } |
064af421 BP |
3332 | } |
3333 | ||
3334 | queue_tx(msg, ofconn, ofconn->reply_counter); | |
3335 | return 0; | |
3336 | } | |
3337 | ||
01149cfd | 3338 | /* Obtains statistic counters for 'rule' within 'p' and stores them into |
bcf84111 BP |
3339 | * '*packet_countp' and '*byte_countp'. The returned statistics include |
3340 | * statistics for all of 'rule''s facets. */ | |
064af421 BP |
3341 | static void |
3342 | query_stats(struct ofproto *p, struct rule *rule, | |
3343 | uint64_t *packet_countp, uint64_t *byte_countp) | |
3344 | { | |
3345 | uint64_t packet_count, byte_count; | |
bcf84111 | 3346 | struct facet *facet; |
064af421 BP |
3347 | struct odp_flow *odp_flows; |
3348 | size_t n_odp_flows; | |
3349 | ||
01149cfd | 3350 | /* Start from historical data for 'rule' itself that are no longer tracked |
bcf84111 | 3351 | * by the datapath. This counts, for example, facets that have expired. */ |
b3137fe8 JG |
3352 | packet_count = rule->packet_count; |
3353 | byte_count = rule->byte_count; | |
3354 | ||
bcf84111 | 3355 | /* Prepare to ask the datapath for statistics on all of the rule's facets. |
01149cfd BP |
3356 | * |
3357 | * Also, add any statistics that are not tracked by the datapath for each | |
bcf84111 | 3358 | * facet. This includes, for example, statistics for packets that were |
01149cfd | 3359 | * executed "by hand" by ofproto via dpif_execute() but must be accounted |
bcf84111 BP |
3360 | * to a rule. */ |
3361 | odp_flows = xzalloc(list_size(&rule->facets) * sizeof *odp_flows); | |
3362 | n_odp_flows = 0; | |
3363 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
3364 | struct odp_flow *odp_flow = &odp_flows[n_odp_flows++]; | |
3365 | odp_flow_key_from_flow(&odp_flow->key, &facet->flow); | |
3366 | packet_count += facet->packet_count; | |
3367 | byte_count += facet->byte_count; | |
064af421 BP |
3368 | } |
3369 | ||
28998b22 | 3370 | /* Fetch up-to-date statistics from the datapath and add them in. */ |
c228a364 | 3371 | if (!dpif_flow_get_multiple(p->dpif, odp_flows, n_odp_flows)) { |
064af421 | 3372 | size_t i; |
bcf84111 | 3373 | |
064af421 BP |
3374 | for (i = 0; i < n_odp_flows; i++) { |
3375 | struct odp_flow *odp_flow = &odp_flows[i]; | |
3376 | packet_count += odp_flow->stats.n_packets; | |
3377 | byte_count += odp_flow->stats.n_bytes; | |
3378 | } | |
3379 | } | |
3380 | free(odp_flows); | |
3381 | ||
01149cfd | 3382 | /* Return the stats to the caller. */ |
064af421 BP |
3383 | *packet_countp = packet_count; |
3384 | *byte_countp = byte_count; | |
3385 | } | |
3386 | ||
c6ebb8fb BP |
3387 | static void |
3388 | calc_flow_duration(long long int start, ovs_be32 *sec, ovs_be32 *nsec) | |
3389 | { | |
3390 | long long int msecs = time_msec() - start; | |
3391 | *sec = htonl(msecs / 1000); | |
3392 | *nsec = htonl((msecs % 1000) * (1000 * 1000)); | |
3393 | } | |
3394 | ||
064af421 | 3395 | static void |
5ecc9d81 BP |
3396 | put_ofp_flow_stats(struct ofconn *ofconn, struct rule *rule, |
3397 | ovs_be16 out_port, struct ofpbuf **replyp) | |
064af421 | 3398 | { |
064af421 BP |
3399 | struct ofp_flow_stats *ofs; |
3400 | uint64_t packet_count, byte_count; | |
3401 | size_t act_len, len; | |
3402 | ||
5ecc9d81 | 3403 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { |
064af421 BP |
3404 | return; |
3405 | } | |
3406 | ||
3407 | act_len = sizeof *rule->actions * rule->n_actions; | |
3408 | len = offsetof(struct ofp_flow_stats, actions) + act_len; | |
3409 | ||
5ecc9d81 | 3410 | query_stats(ofconn->ofproto, rule, &packet_count, &byte_count); |
064af421 | 3411 | |
5ecc9d81 | 3412 | ofs = append_ofp_stats_reply(len, ofconn, replyp); |
064af421 | 3413 | ofs->length = htons(len); |
ad828225 | 3414 | ofs->table_id = 0; |
064af421 | 3415 | ofs->pad = 0; |
d8ae4d67 | 3416 | ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofs->match); |
c6ebb8fb | 3417 | calc_flow_duration(rule->created, &ofs->duration_sec, &ofs->duration_nsec); |
39997502 | 3418 | ofs->cookie = rule->flow_cookie; |
064af421 BP |
3419 | ofs->priority = htons(rule->cr.priority); |
3420 | ofs->idle_timeout = htons(rule->idle_timeout); | |
3421 | ofs->hard_timeout = htons(rule->hard_timeout); | |
39997502 | 3422 | memset(ofs->pad2, 0, sizeof ofs->pad2); |
064af421 BP |
3423 | ofs->packet_count = htonll(packet_count); |
3424 | ofs->byte_count = htonll(byte_count); | |
3dffcf07 BP |
3425 | if (rule->n_actions > 0) { |
3426 | memcpy(ofs->actions, rule->actions, act_len); | |
3427 | } | |
064af421 BP |
3428 | } |
3429 | ||
3c4486a5 BP |
3430 | static bool |
3431 | is_valid_table(uint8_t table_id) | |
064af421 | 3432 | { |
3c4486a5 | 3433 | return table_id == 0 || table_id == 0xff; |
064af421 BP |
3434 | } |
3435 | ||
3436 | static int | |
3269c562 BP |
3437 | handle_flow_stats_request(struct ofconn *ofconn, |
3438 | const struct ofp_stats_request *osr, size_t arg_size) | |
064af421 BP |
3439 | { |
3440 | struct ofp_flow_stats_request *fsr; | |
5ecc9d81 | 3441 | struct ofpbuf *reply; |
064af421 BP |
3442 | |
3443 | if (arg_size != sizeof *fsr) { | |
49bdc010 | 3444 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
3445 | } |
3446 | fsr = (struct ofp_flow_stats_request *) osr->body; | |
3447 | ||
3448 | COVERAGE_INC(ofproto_flows_req); | |
5ecc9d81 | 3449 | reply = start_ofp_stats_reply(osr, 1024); |
3c4486a5 | 3450 | if (is_valid_table(fsr->table_id)) { |
5ecc9d81 | 3451 | struct cls_cursor cursor; |
3c4486a5 | 3452 | struct cls_rule target; |
5ecc9d81 | 3453 | struct rule *rule; |
3c4486a5 | 3454 | |
d8ae4d67 BP |
3455 | ofputil_cls_rule_from_match(&fsr->match, 0, NXFF_OPENFLOW10, 0, |
3456 | &target); | |
5ecc9d81 BP |
3457 | cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); |
3458 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3459 | put_ofp_flow_stats(ofconn, rule, fsr->out_port, &reply); | |
3460 | } | |
3c4486a5 | 3461 | } |
5ecc9d81 | 3462 | queue_tx(reply, ofconn, ofconn->reply_counter); |
3c4486a5 | 3463 | |
064af421 BP |
3464 | return 0; |
3465 | } | |
3466 | ||
09246b99 | 3467 | static void |
5ecc9d81 BP |
3468 | put_nx_flow_stats(struct ofconn *ofconn, struct rule *rule, |
3469 | ovs_be16 out_port, struct ofpbuf **replyp) | |
09246b99 | 3470 | { |
09246b99 BP |
3471 | struct nx_flow_stats *nfs; |
3472 | uint64_t packet_count, byte_count; | |
3473 | size_t act_len, start_len; | |
5ecc9d81 | 3474 | struct ofpbuf *reply; |
09246b99 | 3475 | |
5ecc9d81 | 3476 | if (rule_is_hidden(rule) || !rule_has_out_port(rule, out_port)) { |
09246b99 BP |
3477 | return; |
3478 | } | |
3479 | ||
5ecc9d81 | 3480 | query_stats(ofconn->ofproto, rule, &packet_count, &byte_count); |
09246b99 BP |
3481 | |
3482 | act_len = sizeof *rule->actions * rule->n_actions; | |
3483 | ||
5ecc9d81 BP |
3484 | start_len = (*replyp)->size; |
3485 | append_nxstats_reply(sizeof *nfs + NXM_MAX_LEN + act_len, ofconn, replyp); | |
3486 | reply = *replyp; | |
3487 | ||
3488 | nfs = ofpbuf_put_uninit(reply, sizeof *nfs); | |
09246b99 BP |
3489 | nfs->table_id = 0; |
3490 | nfs->pad = 0; | |
3491 | calc_flow_duration(rule->created, &nfs->duration_sec, &nfs->duration_nsec); | |
3492 | nfs->cookie = rule->flow_cookie; | |
3493 | nfs->priority = htons(rule->cr.priority); | |
3494 | nfs->idle_timeout = htons(rule->idle_timeout); | |
3495 | nfs->hard_timeout = htons(rule->hard_timeout); | |
5ecc9d81 | 3496 | nfs->match_len = htons(nx_put_match(reply, &rule->cr)); |
09246b99 BP |
3497 | memset(nfs->pad2, 0, sizeof nfs->pad2); |
3498 | nfs->packet_count = htonll(packet_count); | |
3499 | nfs->byte_count = htonll(byte_count); | |
3500 | if (rule->n_actions > 0) { | |
5ecc9d81 | 3501 | ofpbuf_put(reply, rule->actions, act_len); |
09246b99 | 3502 | } |
5ecc9d81 | 3503 | nfs->length = htons(reply->size - start_len); |
09246b99 BP |
3504 | } |
3505 | ||
3506 | static int | |
3507 | handle_nxst_flow(struct ofconn *ofconn, struct ofpbuf *b) | |
3508 | { | |
3509 | struct nx_flow_stats_request *nfsr; | |
09246b99 | 3510 | struct cls_rule target; |
5ecc9d81 | 3511 | struct ofpbuf *reply; |
09246b99 BP |
3512 | int error; |
3513 | ||
3514 | /* Dissect the message. */ | |
3515 | nfsr = ofpbuf_try_pull(b, sizeof *nfsr); | |
3516 | if (!nfsr) { | |
3517 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3518 | } | |
3519 | error = nx_pull_match(b, ntohs(nfsr->match_len), 0, &target); | |
3520 | if (error) { | |
3521 | return error; | |
3522 | } | |
3523 | ||
3524 | COVERAGE_INC(ofproto_flows_req); | |
5ecc9d81 | 3525 | reply = start_nxstats_reply(&nfsr->nsm, 1024); |
3c4486a5 | 3526 | if (is_valid_table(nfsr->table_id)) { |
5ecc9d81 BP |
3527 | struct cls_cursor cursor; |
3528 | struct rule *rule; | |
3529 | ||
3530 | cls_cursor_init(&cursor, &ofconn->ofproto->cls, &target); | |
3531 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3532 | put_nx_flow_stats(ofconn, rule, nfsr->out_port, &reply); | |
3533 | } | |
3c4486a5 | 3534 | } |
5ecc9d81 BP |
3535 | queue_tx(reply, ofconn, ofconn->reply_counter); |
3536 | ||
09246b99 BP |
3537 | return 0; |
3538 | } | |
3539 | ||
4f2cad2c | 3540 | static void |
5ecc9d81 | 3541 | flow_stats_ds(struct ofproto *ofproto, struct rule *rule, struct ds *results) |
4f2cad2c | 3542 | { |
4f2cad2c JP |
3543 | struct ofp_match match; |
3544 | uint64_t packet_count, byte_count; | |
3545 | size_t act_len = sizeof *rule->actions * rule->n_actions; | |
3546 | ||
5ecc9d81 | 3547 | query_stats(ofproto, rule, &packet_count, &byte_count); |
d8ae4d67 | 3548 | ofputil_cls_rule_to_match(&rule->cr, NXFF_OPENFLOW10, &match); |
4f2cad2c JP |
3549 | |
3550 | ds_put_format(results, "duration=%llds, ", | |
3551 | (time_msec() - rule->created) / 1000); | |
52ae00b3 | 3552 | ds_put_format(results, "priority=%u, ", rule->cr.priority); |
4f2cad2c JP |
3553 | ds_put_format(results, "n_packets=%"PRIu64", ", packet_count); |
3554 | ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count); | |
3555 | ofp_print_match(results, &match, true); | |
3dffcf07 BP |
3556 | if (act_len > 0) { |
3557 | ofp_print_actions(results, &rule->actions->header, act_len); | |
3c8552c1 JP |
3558 | } else { |
3559 | ds_put_cstr(results, "drop"); | |
3dffcf07 | 3560 | } |
4f2cad2c JP |
3561 | ds_put_cstr(results, "\n"); |
3562 | } | |
3563 | ||
d295e8e9 | 3564 | /* Adds a pretty-printed description of all flows to 'results', including |
4f2cad2c JP |
3565 | * those marked hidden by secchan (e.g., by in-band control). */ |
3566 | void | |
3567 | ofproto_get_all_flows(struct ofproto *p, struct ds *results) | |
3568 | { | |
5ecc9d81 BP |
3569 | struct cls_cursor cursor; |
3570 | struct rule *rule; | |
064af421 | 3571 | |
5ecc9d81 BP |
3572 | cls_cursor_init(&cursor, &p->cls, NULL); |
3573 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3574 | flow_stats_ds(p, rule, results); | |
064af421 | 3575 | } |
064af421 BP |
3576 | } |
3577 | ||
27d34fce BP |
3578 | static void |
3579 | query_aggregate_stats(struct ofproto *ofproto, struct cls_rule *target, | |
734bbeb4 | 3580 | ovs_be16 out_port, uint8_t table_id, |
27d34fce BP |
3581 | struct ofp_aggregate_stats_reply *oasr) |
3582 | { | |
5ecc9d81 BP |
3583 | uint64_t total_packets = 0; |
3584 | uint64_t total_bytes = 0; | |
3585 | int n_flows = 0; | |
27d34fce BP |
3586 | |
3587 | COVERAGE_INC(ofproto_agg_request); | |
5ecc9d81 | 3588 | |
3c4486a5 | 3589 | if (is_valid_table(table_id)) { |
5ecc9d81 BP |
3590 | struct cls_cursor cursor; |
3591 | struct rule *rule; | |
3c4486a5 | 3592 | |
5ecc9d81 BP |
3593 | cls_cursor_init(&cursor, &ofproto->cls, target); |
3594 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3595 | if (!rule_is_hidden(rule) && rule_has_out_port(rule, out_port)) { | |
3596 | uint64_t packet_count; | |
3597 | uint64_t byte_count; | |
3598 | ||
3599 | query_stats(ofproto, rule, &packet_count, &byte_count); | |
3600 | ||
3601 | total_packets += packet_count; | |
3602 | total_bytes += byte_count; | |
3603 | n_flows++; | |
3604 | } | |
3605 | } | |
3c4486a5 | 3606 | } |
27d34fce | 3607 | |
5ecc9d81 BP |
3608 | oasr->flow_count = htonl(n_flows); |
3609 | oasr->packet_count = htonll(total_packets); | |
3610 | oasr->byte_count = htonll(total_bytes); | |
27d34fce BP |
3611 | memset(oasr->pad, 0, sizeof oasr->pad); |
3612 | } | |
3613 | ||
064af421 | 3614 | static int |
3269c562 | 3615 | handle_aggregate_stats_request(struct ofconn *ofconn, |
064af421 BP |
3616 | const struct ofp_stats_request *osr, |
3617 | size_t arg_size) | |
3618 | { | |
27d34fce | 3619 | struct ofp_aggregate_stats_request *request; |
064af421 | 3620 | struct ofp_aggregate_stats_reply *reply; |
064af421 BP |
3621 | struct cls_rule target; |
3622 | struct ofpbuf *msg; | |
3623 | ||
27d34fce | 3624 | if (arg_size != sizeof *request) { |
49bdc010 | 3625 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 | 3626 | } |
27d34fce | 3627 | request = (struct ofp_aggregate_stats_request *) osr->body; |
064af421 | 3628 | |
d8ae4d67 BP |
3629 | ofputil_cls_rule_from_match(&request->match, 0, NXFF_OPENFLOW10, 0, |
3630 | &target); | |
064af421 | 3631 | |
06a5e131 BP |
3632 | msg = start_ofp_stats_reply(osr, sizeof *reply); |
3633 | reply = append_ofp_stats_reply(sizeof *reply, ofconn, &msg); | |
27d34fce BP |
3634 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, |
3635 | request->table_id, reply); | |
064af421 BP |
3636 | queue_tx(msg, ofconn, ofconn->reply_counter); |
3637 | return 0; | |
3638 | } | |
3639 | ||
09246b99 BP |
3640 | static int |
3641 | handle_nxst_aggregate(struct ofconn *ofconn, struct ofpbuf *b) | |
3642 | { | |
3643 | struct nx_aggregate_stats_request *request; | |
3644 | struct ofp_aggregate_stats_reply *reply; | |
3645 | struct cls_rule target; | |
3646 | struct ofpbuf *buf; | |
3647 | int error; | |
3648 | ||
3649 | /* Dissect the message. */ | |
3650 | request = ofpbuf_try_pull(b, sizeof *request); | |
3651 | if (!request) { | |
3652 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3653 | } | |
3654 | error = nx_pull_match(b, ntohs(request->match_len), 0, &target); | |
3655 | if (error) { | |
3656 | return error; | |
3657 | } | |
3658 | ||
3659 | /* Reply. */ | |
3660 | COVERAGE_INC(ofproto_flows_req); | |
3661 | buf = start_nxstats_reply(&request->nsm, sizeof *reply); | |
3662 | reply = ofpbuf_put_uninit(buf, sizeof *reply); | |
3663 | query_aggregate_stats(ofconn->ofproto, &target, request->out_port, | |
3664 | request->table_id, reply); | |
3665 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
3666 | ||
3667 | return 0; | |
3668 | } | |
3669 | ||
c1c9c9c4 BP |
3670 | struct queue_stats_cbdata { |
3671 | struct ofconn *ofconn; | |
ca0f572c | 3672 | struct ofport *ofport; |
c1c9c9c4 | 3673 | struct ofpbuf *msg; |
c1c9c9c4 BP |
3674 | }; |
3675 | ||
3676 | static void | |
db9220c3 | 3677 | put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id, |
c1c9c9c4 BP |
3678 | const struct netdev_queue_stats *stats) |
3679 | { | |
3680 | struct ofp_queue_stats *reply; | |
3681 | ||
06a5e131 | 3682 | reply = append_ofp_stats_reply(sizeof *reply, cbdata->ofconn, &cbdata->msg); |
ca0f572c | 3683 | reply->port_no = htons(cbdata->ofport->opp.port_no); |
c1c9c9c4 BP |
3684 | memset(reply->pad, 0, sizeof reply->pad); |
3685 | reply->queue_id = htonl(queue_id); | |
3686 | reply->tx_bytes = htonll(stats->tx_bytes); | |
3687 | reply->tx_packets = htonll(stats->tx_packets); | |
3688 | reply->tx_errors = htonll(stats->tx_errors); | |
3689 | } | |
3690 | ||
3691 | static void | |
db9220c3 | 3692 | handle_queue_stats_dump_cb(uint32_t queue_id, |
c1c9c9c4 BP |
3693 | struct netdev_queue_stats *stats, |
3694 | void *cbdata_) | |
3695 | { | |
3696 | struct queue_stats_cbdata *cbdata = cbdata_; | |
3697 | ||
3698 | put_queue_stats(cbdata, queue_id, stats); | |
3699 | } | |
3700 | ||
3701 | static void | |
ca0f572c | 3702 | handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id, |
c1c9c9c4 BP |
3703 | struct queue_stats_cbdata *cbdata) |
3704 | { | |
ca0f572c | 3705 | cbdata->ofport = port; |
c1c9c9c4 BP |
3706 | if (queue_id == OFPQ_ALL) { |
3707 | netdev_dump_queue_stats(port->netdev, | |
3708 | handle_queue_stats_dump_cb, cbdata); | |
3709 | } else { | |
3710 | struct netdev_queue_stats stats; | |
3711 | ||
1ac788f6 BP |
3712 | if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) { |
3713 | put_queue_stats(cbdata, queue_id, &stats); | |
3714 | } | |
c1c9c9c4 BP |
3715 | } |
3716 | } | |
3717 | ||
3718 | static int | |
3269c562 | 3719 | handle_queue_stats_request(struct ofconn *ofconn, |
c1c9c9c4 BP |
3720 | const struct ofp_stats_request *osr, |
3721 | size_t arg_size) | |
3722 | { | |
3269c562 | 3723 | struct ofproto *ofproto = ofconn->ofproto; |
c1c9c9c4 BP |
3724 | struct ofp_queue_stats_request *qsr; |
3725 | struct queue_stats_cbdata cbdata; | |
3726 | struct ofport *port; | |
3727 | unsigned int port_no; | |
3728 | uint32_t queue_id; | |
3729 | ||
3730 | if (arg_size != sizeof *qsr) { | |
3731 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3732 | } | |
3733 | qsr = (struct ofp_queue_stats_request *) osr->body; | |
3734 | ||
3735 | COVERAGE_INC(ofproto_queue_req); | |
3736 | ||
3737 | cbdata.ofconn = ofconn; | |
06a5e131 | 3738 | cbdata.msg = start_ofp_stats_reply(osr, 128); |
c1c9c9c4 BP |
3739 | |
3740 | port_no = ntohs(qsr->port_no); | |
3741 | queue_id = ntohl(qsr->queue_id); | |
3742 | if (port_no == OFPP_ALL) { | |
4e8e4213 | 3743 | HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) { |
ca0f572c | 3744 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3745 | } |
3746 | } else if (port_no < ofproto->max_ports) { | |
ca0f572c | 3747 | port = get_port(ofproto, ofp_port_to_odp_port(port_no)); |
c1c9c9c4 | 3748 | if (port) { |
ca0f572c | 3749 | handle_queue_stats_for_port(port, queue_id, &cbdata); |
c1c9c9c4 BP |
3750 | } |
3751 | } else { | |
3752 | ofpbuf_delete(cbdata.msg); | |
3753 | return ofp_mkerr(OFPET_QUEUE_OP_FAILED, OFPQOFC_BAD_PORT); | |
3754 | } | |
3755 | queue_tx(cbdata.msg, ofconn, ofconn->reply_counter); | |
3756 | ||
3757 | return 0; | |
3758 | } | |
3759 | ||
09246b99 BP |
3760 | static int |
3761 | handle_vendor_stats_request(struct ofconn *ofconn, | |
3762 | struct ofp_stats_request *osr, size_t arg_size) | |
3763 | { | |
3764 | struct nicira_stats_msg *nsm; | |
3765 | struct ofpbuf b; | |
3766 | ovs_be32 vendor; | |
3767 | ||
3768 | if (arg_size < 4) { | |
3769 | VLOG_WARN_RL(&rl, "truncated vendor stats request body"); | |
3770 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3771 | } | |
3772 | ||
3773 | memcpy(&vendor, osr->body, sizeof vendor); | |
3774 | if (vendor != htonl(NX_VENDOR_ID)) { | |
3775 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
3776 | } | |
3777 | ||
487cedcf | 3778 | if (ntohs(osr->header.length) < sizeof(struct nicira_stats_msg)) { |
09246b99 BP |
3779 | VLOG_WARN_RL(&rl, "truncated Nicira stats request"); |
3780 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
3781 | } | |
3782 | ||
3783 | nsm = (struct nicira_stats_msg *) osr; | |
3784 | b.data = nsm; | |
3785 | b.size = ntohs(nsm->header.length); | |
3786 | switch (ntohl(nsm->subtype)) { | |
3787 | case NXST_FLOW: | |
3788 | return handle_nxst_flow(ofconn, &b); | |
3789 | ||
3790 | case NXST_AGGREGATE: | |
3791 | return handle_nxst_aggregate(ofconn, &b); | |
3792 | ||
3793 | default: | |
3794 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
3795 | } | |
3796 | } | |
3797 | ||
064af421 | 3798 | static int |
3269c562 | 3799 | handle_stats_request(struct ofconn *ofconn, struct ofp_header *oh) |
064af421 BP |
3800 | { |
3801 | struct ofp_stats_request *osr; | |
3802 | size_t arg_size; | |
3803 | int error; | |
3804 | ||
3805 | error = check_ofp_message_array(oh, OFPT_STATS_REQUEST, sizeof *osr, | |
3806 | 1, &arg_size); | |
3807 | if (error) { | |
3808 | return error; | |
3809 | } | |
3810 | osr = (struct ofp_stats_request *) oh; | |
3811 | ||
3812 | switch (ntohs(osr->type)) { | |
3813 | case OFPST_DESC: | |
3269c562 | 3814 | return handle_desc_stats_request(ofconn, osr); |
064af421 BP |
3815 | |
3816 | case OFPST_FLOW: | |
3269c562 | 3817 | return handle_flow_stats_request(ofconn, osr, arg_size); |
064af421 BP |
3818 | |
3819 | case OFPST_AGGREGATE: | |
3269c562 | 3820 | return handle_aggregate_stats_request(ofconn, osr, arg_size); |
064af421 BP |
3821 | |
3822 | case OFPST_TABLE: | |
3269c562 | 3823 | return handle_table_stats_request(ofconn, osr); |
064af421 BP |
3824 | |
3825 | case OFPST_PORT: | |
3269c562 | 3826 | return handle_port_stats_request(ofconn, osr, arg_size); |
064af421 | 3827 | |
c1c9c9c4 | 3828 | case OFPST_QUEUE: |
3269c562 | 3829 | return handle_queue_stats_request(ofconn, osr, arg_size); |
c1c9c9c4 | 3830 | |
064af421 | 3831 | case OFPST_VENDOR: |
09246b99 | 3832 | return handle_vendor_stats_request(ofconn, osr, arg_size); |
064af421 BP |
3833 | |
3834 | default: | |
3835 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_STAT); | |
3836 | } | |
3837 | } | |
3838 | ||
3839 | static long long int | |
3840 | msec_from_nsec(uint64_t sec, uint32_t nsec) | |
3841 | { | |
3842 | return !sec ? 0 : sec * 1000 + nsec / 1000000; | |
3843 | } | |
3844 | ||
3845 | static void | |
bcf84111 BP |
3846 | facet_update_time(struct ofproto *ofproto, struct facet *facet, |
3847 | const struct odp_flow_stats *stats) | |
064af421 BP |
3848 | { |
3849 | long long int used = msec_from_nsec(stats->used_sec, stats->used_nsec); | |
bcf84111 BP |
3850 | if (used > facet->used) { |
3851 | facet->used = used; | |
3852 | if (used > facet->rule->used) { | |
3853 | facet->rule->used = used; | |
4836f9f2 | 3854 | } |
bcf84111 | 3855 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); |
064af421 BP |
3856 | } |
3857 | } | |
3858 | ||
bcf84111 BP |
3859 | /* Folds the statistics from 'stats' into the counters in 'facet'. |
3860 | * | |
3861 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
3862 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
3863 | * packet that was sent by hand or if it represents statistics that have been | |
3864 | * cleared out of the datapath. */ | |
064af421 | 3865 | static void |
bcf84111 BP |
3866 | facet_update_stats(struct ofproto *ofproto, struct facet *facet, |
3867 | const struct odp_flow_stats *stats) | |
064af421 | 3868 | { |
064af421 | 3869 | if (stats->n_packets) { |
bcf84111 BP |
3870 | facet_update_time(ofproto, facet, stats); |
3871 | facet->packet_count += stats->n_packets; | |
3872 | facet->byte_count += stats->n_bytes; | |
3873 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
064af421 BP |
3874 | } |
3875 | } | |
3876 | ||
3052b0c5 BP |
3877 | struct flow_mod { |
3878 | struct cls_rule cr; | |
3879 | ovs_be64 cookie; | |
3880 | uint16_t command; | |
3881 | uint16_t idle_timeout; | |
3882 | uint16_t hard_timeout; | |
3883 | uint32_t buffer_id; | |
3884 | uint16_t out_port; | |
3885 | uint16_t flags; | |
3886 | union ofp_action *actions; | |
3887 | size_t n_actions; | |
3888 | }; | |
3889 | ||
79eee1eb BP |
3890 | /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT |
3891 | * in which no matching flow already exists in the flow table. | |
3892 | * | |
3893 | * Adds the flow specified by 'ofm', which is followed by 'n_actions' | |
3269c562 BP |
3894 | * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an |
3895 | * OpenFlow error code as encoded by ofp_mkerr() on failure. | |
79eee1eb BP |
3896 | * |
3897 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3898 | * if any. */ | |
064af421 | 3899 | static int |
3052b0c5 | 3900 | add_flow(struct ofconn *ofconn, struct flow_mod *fm) |
064af421 | 3901 | { |
3269c562 | 3902 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
3903 | struct ofpbuf *packet; |
3904 | struct rule *rule; | |
3905 | uint16_t in_port; | |
3906 | int error; | |
3907 | ||
3052b0c5 BP |
3908 | if (fm->flags & OFPFF_CHECK_OVERLAP |
3909 | && classifier_rule_overlaps(&p->cls, &fm->cr)) { | |
3910 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_OVERLAP); | |
49bdc010 JP |
3911 | } |
3912 | ||
064af421 | 3913 | error = 0; |
3052b0c5 BP |
3914 | if (fm->buffer_id != UINT32_MAX) { |
3915 | error = pktbuf_retrieve(ofconn->pktbuf, fm->buffer_id, | |
064af421 | 3916 | &packet, &in_port); |
212fe71c BP |
3917 | } else { |
3918 | packet = NULL; | |
165cd8a3 | 3919 | in_port = UINT16_MAX; |
064af421 BP |
3920 | } |
3921 | ||
bcf84111 BP |
3922 | rule = rule_create(&fm->cr, fm->actions, fm->n_actions, |
3923 | fm->idle_timeout, fm->hard_timeout, fm->cookie, | |
3924 | fm->flags & OFPFF_SEND_FLOW_REM); | |
afe75089 BP |
3925 | rule_insert(p, rule); |
3926 | if (packet) { | |
3927 | rule_execute(p, rule, in_port, packet); | |
3928 | } | |
064af421 BP |
3929 | return error; |
3930 | } | |
3931 | ||
79eee1eb | 3932 | static struct rule * |
3052b0c5 | 3933 | find_flow_strict(struct ofproto *p, const struct flow_mod *fm) |
064af421 | 3934 | { |
3052b0c5 | 3935 | return rule_from_cls_rule(classifier_find_rule_exactly(&p->cls, &fm->cr)); |
79eee1eb | 3936 | } |
064af421 | 3937 | |
79eee1eb | 3938 | static int |
3269c562 | 3939 | send_buffered_packet(struct ofconn *ofconn, |
3052b0c5 | 3940 | struct rule *rule, uint32_t buffer_id) |
79eee1eb BP |
3941 | { |
3942 | struct ofpbuf *packet; | |
3943 | uint16_t in_port; | |
79eee1eb | 3944 | int error; |
064af421 | 3945 | |
3052b0c5 | 3946 | if (buffer_id == UINT32_MAX) { |
79eee1eb | 3947 | return 0; |
064af421 | 3948 | } |
79eee1eb | 3949 | |
3052b0c5 | 3950 | error = pktbuf_retrieve(ofconn->pktbuf, buffer_id, &packet, &in_port); |
79eee1eb BP |
3951 | if (error) { |
3952 | return error; | |
3953 | } | |
3954 | ||
bcf84111 | 3955 | rule_execute(ofconn->ofproto, rule, in_port, packet); |
79eee1eb | 3956 | |
064af421 BP |
3957 | return 0; |
3958 | } | |
79eee1eb BP |
3959 | \f |
3960 | /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */ | |
064af421 BP |
3961 | |
3962 | struct modify_flows_cbdata { | |
3963 | struct ofproto *ofproto; | |
3052b0c5 | 3964 | const struct flow_mod *fm; |
79eee1eb | 3965 | struct rule *match; |
064af421 BP |
3966 | }; |
3967 | ||
3052b0c5 BP |
3968 | static int modify_flow(struct ofproto *, const struct flow_mod *, |
3969 | struct rule *); | |
79eee1eb BP |
3970 | |
3971 | /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as | |
3972 | * encoded by ofp_mkerr() on failure. | |
3973 | * | |
3974 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
3975 | * if any. */ | |
3976 | static int | |
3052b0c5 | 3977 | modify_flows_loose(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 3978 | { |
5ecc9d81 BP |
3979 | struct ofproto *p = ofconn->ofproto; |
3980 | struct rule *match = NULL; | |
3981 | struct cls_cursor cursor; | |
3982 | struct rule *rule; | |
79eee1eb | 3983 | |
5ecc9d81 BP |
3984 | cls_cursor_init(&cursor, &p->cls, &fm->cr); |
3985 | CLS_CURSOR_FOR_EACH (rule, cr, &cursor) { | |
3986 | if (!rule_is_hidden(rule)) { | |
3987 | match = rule; | |
3988 | modify_flow(p, fm, rule); | |
3989 | } | |
3990 | } | |
79eee1eb | 3991 | |
5ecc9d81 | 3992 | if (match) { |
d6302b0f BP |
3993 | /* This credits the packet to whichever flow happened to match last. |
3994 | * That's weird. Maybe we should do a lookup for the flow that | |
3995 | * actually matches the packet? Who knows. */ | |
5ecc9d81 | 3996 | send_buffered_packet(ofconn, match, fm->buffer_id); |
79eee1eb BP |
3997 | return 0; |
3998 | } else { | |
3052b0c5 | 3999 | return add_flow(ofconn, fm); |
79eee1eb BP |
4000 | } |
4001 | } | |
4002 | ||
4003 | /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error | |
4004 | * code as encoded by ofp_mkerr() on failure. | |
4005 | * | |
4006 | * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id, | |
4007 | * if any. */ | |
4008 | static int | |
3052b0c5 | 4009 | modify_flow_strict(struct ofconn *ofconn, struct flow_mod *fm) |
79eee1eb | 4010 | { |
3052b0c5 BP |
4011 | struct ofproto *p = ofconn->ofproto; |
4012 | struct rule *rule = find_flow_strict(p, fm); | |
79eee1eb | 4013 | if (rule && !rule_is_hidden(rule)) { |
3052b0c5 BP |
4014 | modify_flow(p, fm, rule); |
4015 | return send_buffered_packet(ofconn, rule, fm->buffer_id); | |
79eee1eb | 4016 | } else { |
3052b0c5 | 4017 | return add_flow(ofconn, fm); |
79eee1eb BP |
4018 | } |
4019 | } | |
4020 | ||
79eee1eb BP |
4021 | /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has |
4022 | * been identified as a flow in 'p''s flow table to be modified, by changing | |
4023 | * the rule's actions to match those in 'ofm' (which is followed by 'n_actions' | |
4024 | * ofp_action[] structures). */ | |
064af421 | 4025 | static int |
3052b0c5 | 4026 | modify_flow(struct ofproto *p, const struct flow_mod *fm, struct rule *rule) |
064af421 | 4027 | { |
3052b0c5 | 4028 | size_t actions_len = fm->n_actions * sizeof *rule->actions; |
79eee1eb | 4029 | |
3052b0c5 | 4030 | rule->flow_cookie = fm->cookie; |
79eee1eb BP |
4031 | |
4032 | /* If the actions are the same, do nothing. */ | |
3052b0c5 BP |
4033 | if (fm->n_actions == rule->n_actions |
4034 | && (!fm->n_actions | |
4035 | || !memcmp(fm->actions, rule->actions, actions_len))) { | |
79eee1eb BP |
4036 | return 0; |
4037 | } | |
4038 | ||
4039 | /* Replace actions. */ | |
4040 | free(rule->actions); | |
3052b0c5 BP |
4041 | rule->actions = fm->n_actions ? xmemdup(fm->actions, actions_len) : NULL; |
4042 | rule->n_actions = fm->n_actions; | |
79eee1eb | 4043 | |
bcf84111 | 4044 | p->need_revalidate = true; |
79eee1eb BP |
4045 | |
4046 | return 0; | |
4047 | } | |
4048 | \f | |
4049 | /* OFPFC_DELETE implementation. */ | |
4050 | ||
8054fc48 | 4051 | static void delete_flow(struct ofproto *, struct rule *, ovs_be16 out_port); |
79eee1eb BP |
4052 | |
4053 | /* Implements OFPFC_DELETE. */ | |
4054 | static void | |
3052b0c5 | 4055 | delete_flows_loose(struct ofproto *p, const struct flow_mod *fm) |
79eee1eb | 4056 | { |
5ecc9d81 BP |
4057 | struct rule *rule, *next_rule; |
4058 | struct cls_cursor cursor; | |
064af421 | 4059 | |
5ecc9d81 BP |
4060 | cls_cursor_init(&cursor, &p->cls, &fm->cr); |
4061 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
4062 | delete_flow(p, rule, htons(fm->out_port)); | |
4063 | } | |
064af421 BP |
4064 | } |
4065 | ||
79eee1eb BP |
4066 | /* Implements OFPFC_DELETE_STRICT. */ |
4067 | static void | |
3052b0c5 | 4068 | delete_flow_strict(struct ofproto *p, struct flow_mod *fm) |
79eee1eb | 4069 | { |
3052b0c5 | 4070 | struct rule *rule = find_flow_strict(p, fm); |
79eee1eb | 4071 | if (rule) { |
3052b0c5 | 4072 | delete_flow(p, rule, htons(fm->out_port)); |
79eee1eb BP |
4073 | } |
4074 | } | |
4075 | ||
79eee1eb BP |
4076 | /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has |
4077 | * been identified as a flow to delete from 'p''s flow table, by deleting the | |
4078 | * flow and sending out a OFPT_FLOW_REMOVED message to any interested | |
4079 | * controller. | |
4080 | * | |
4081 | * Will not delete 'rule' if it is hidden. Will delete 'rule' only if | |
4082 | * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the | |
4083 | * specified 'out_port'. */ | |
4084 | static void | |
8054fc48 | 4085 | delete_flow(struct ofproto *p, struct rule *rule, ovs_be16 out_port) |
79eee1eb BP |
4086 | { |
4087 | if (rule_is_hidden(rule)) { | |
4088 | return; | |
4089 | } | |
4090 | ||
4091 | if (out_port != htons(OFPP_NONE) && !rule_has_out_port(rule, out_port)) { | |
4092 | return; | |
4093 | } | |
4094 | ||
bcf84111 | 4095 | rule_send_removed(p, rule, OFPRR_DELETE); |
79eee1eb BP |
4096 | rule_remove(p, rule); |
4097 | } | |
4098 | \f | |
064af421 | 4099 | static int |
3052b0c5 | 4100 | flow_mod_core(struct ofconn *ofconn, struct flow_mod *fm) |
064af421 | 4101 | { |
3052b0c5 | 4102 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
4103 | int error; |
4104 | ||
3052b0c5 | 4105 | error = reject_slave_controller(ofconn, "flow_mod"); |
9deba63b BP |
4106 | if (error) { |
4107 | return error; | |
4108 | } | |
3052b0c5 | 4109 | |
f1defbf9 BP |
4110 | error = validate_actions(fm->actions, fm->n_actions, |
4111 | &fm->cr.flow, p->max_ports); | |
064af421 BP |
4112 | if (error) { |
4113 | return error; | |
4114 | } | |
4115 | ||
49bdc010 JP |
4116 | /* We do not support the emergency flow cache. It will hopefully |
4117 | * get dropped from OpenFlow in the near future. */ | |
3052b0c5 | 4118 | if (fm->flags & OFPFF_EMERG) { |
49bdc010 JP |
4119 | /* There isn't a good fit for an error code, so just state that the |
4120 | * flow table is full. */ | |
4121 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_ALL_TABLES_FULL); | |
4122 | } | |
4123 | ||
3052b0c5 BP |
4124 | switch (fm->command) { |
4125 | case OFPFC_ADD: | |
4126 | return add_flow(ofconn, fm); | |
4127 | ||
4128 | case OFPFC_MODIFY: | |
4129 | return modify_flows_loose(ofconn, fm); | |
4130 | ||
4131 | case OFPFC_MODIFY_STRICT: | |
4132 | return modify_flow_strict(ofconn, fm); | |
4133 | ||
4134 | case OFPFC_DELETE: | |
4135 | delete_flows_loose(p, fm); | |
4136 | return 0; | |
4137 | ||
4138 | case OFPFC_DELETE_STRICT: | |
4139 | delete_flow_strict(p, fm); | |
4140 | return 0; | |
4141 | ||
4142 | default: | |
4143 | return ofp_mkerr(OFPET_FLOW_MOD_FAILED, OFPFMFC_BAD_COMMAND); | |
4144 | } | |
4145 | } | |
4146 | ||
4147 | static int | |
09246b99 | 4148 | handle_ofpt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) |
3052b0c5 BP |
4149 | { |
4150 | struct ofp_match orig_match; | |
4151 | struct ofp_flow_mod *ofm; | |
4152 | struct flow_mod fm; | |
4153 | struct ofpbuf b; | |
4154 | int error; | |
4155 | ||
4156 | b.data = oh; | |
4157 | b.size = ntohs(oh->length); | |
4158 | ||
4159 | /* Dissect the message. */ | |
4160 | ofm = ofpbuf_try_pull(&b, sizeof *ofm); | |
4161 | if (!ofm) { | |
4162 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4163 | } | |
4164 | error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); | |
4165 | if (error) { | |
4166 | return error; | |
4167 | } | |
4168 | ||
4169 | /* Normalize ofm->match. If normalization actually changes anything, then | |
3f09c339 BP |
4170 | * log the differences. */ |
4171 | ofm->match.pad1[0] = ofm->match.pad2[0] = 0; | |
4172 | orig_match = ofm->match; | |
064af421 | 4173 | normalize_match(&ofm->match); |
3f09c339 BP |
4174 | if (memcmp(&ofm->match, &orig_match, sizeof orig_match)) { |
4175 | static struct vlog_rate_limit normal_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4176 | if (!VLOG_DROP_INFO(&normal_rl)) { | |
4177 | char *old = ofp_match_to_literal_string(&orig_match); | |
4178 | char *new = ofp_match_to_literal_string(&ofm->match); | |
4179 | VLOG_INFO("%s: normalization changed ofp_match, details:", | |
4180 | rconn_get_name(ofconn->rconn)); | |
4181 | VLOG_INFO(" pre: %s", old); | |
4182 | VLOG_INFO("post: %s", new); | |
4183 | free(old); | |
4184 | free(new); | |
4185 | } | |
4186 | } | |
4187 | ||
3052b0c5 | 4188 | /* Translate the message. */ |
d8ae4d67 BP |
4189 | ofputil_cls_rule_from_match(&ofm->match, ntohs(ofm->priority), |
4190 | ofconn->flow_format, ofm->cookie, &fm.cr); | |
3052b0c5 BP |
4191 | fm.cookie = ofm->cookie; |
4192 | fm.command = ntohs(ofm->command); | |
4193 | fm.idle_timeout = ntohs(ofm->idle_timeout); | |
4194 | fm.hard_timeout = ntohs(ofm->hard_timeout); | |
4195 | fm.buffer_id = ntohl(ofm->buffer_id); | |
4196 | fm.out_port = ntohs(ofm->out_port); | |
4197 | fm.flags = ntohs(ofm->flags); | |
064af421 | 4198 | |
3052b0c5 BP |
4199 | /* Execute the command. */ |
4200 | return flow_mod_core(ofconn, &fm); | |
064af421 BP |
4201 | } |
4202 | ||
09246b99 BP |
4203 | static int |
4204 | handle_nxt_flow_mod(struct ofconn *ofconn, struct ofp_header *oh) | |
4205 | { | |
4206 | struct nx_flow_mod *nfm; | |
4207 | struct flow_mod fm; | |
4208 | struct ofpbuf b; | |
4209 | int error; | |
4210 | ||
4211 | b.data = oh; | |
4212 | b.size = ntohs(oh->length); | |
4213 | ||
4214 | /* Dissect the message. */ | |
4215 | nfm = ofpbuf_try_pull(&b, sizeof *nfm); | |
4216 | if (!nfm) { | |
4217 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4218 | } | |
4219 | error = nx_pull_match(&b, ntohs(nfm->match_len), ntohs(nfm->priority), | |
4220 | &fm.cr); | |
4221 | if (error) { | |
4222 | return error; | |
4223 | } | |
4224 | error = ofputil_pull_actions(&b, b.size, &fm.actions, &fm.n_actions); | |
4225 | if (error) { | |
4226 | return error; | |
4227 | } | |
4228 | ||
4229 | /* Translate the message. */ | |
4230 | fm.cookie = nfm->cookie; | |
4231 | fm.command = ntohs(nfm->command); | |
4232 | fm.idle_timeout = ntohs(nfm->idle_timeout); | |
4233 | fm.hard_timeout = ntohs(nfm->hard_timeout); | |
4234 | fm.buffer_id = ntohl(nfm->buffer_id); | |
4235 | fm.out_port = ntohs(nfm->out_port); | |
4236 | fm.flags = ntohs(nfm->flags); | |
4237 | ||
4238 | /* Execute the command. */ | |
4239 | return flow_mod_core(ofconn, &fm); | |
4240 | } | |
4241 | ||
659586ef | 4242 | static int |
b70eac89 | 4243 | handle_tun_id_from_cookie(struct ofconn *ofconn, struct nxt_tun_id_cookie *msg) |
659586ef JG |
4244 | { |
4245 | int error; | |
4246 | ||
4247 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
4248 | if (error) { | |
4249 | return error; | |
4250 | } | |
4251 | ||
b70eac89 | 4252 | ofconn->flow_format = msg->set ? NXFF_TUN_ID_FROM_COOKIE : NXFF_OPENFLOW10; |
659586ef JG |
4253 | return 0; |
4254 | } | |
4255 | ||
9deba63b | 4256 | static int |
3269c562 | 4257 | handle_role_request(struct ofconn *ofconn, struct nicira_header *msg) |
9deba63b BP |
4258 | { |
4259 | struct nx_role_request *nrr; | |
4260 | struct nx_role_request *reply; | |
4261 | struct ofpbuf *buf; | |
4262 | uint32_t role; | |
4263 | ||
4264 | if (ntohs(msg->header.length) != sizeof *nrr) { | |
100e95db | 4265 | VLOG_WARN_RL(&rl, "received role request of length %u (expected %zu)", |
9deba63b BP |
4266 | ntohs(msg->header.length), sizeof *nrr); |
4267 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); | |
4268 | } | |
4269 | nrr = (struct nx_role_request *) msg; | |
4270 | ||
5899143f | 4271 | if (ofconn->type != OFCONN_PRIMARY) { |
9deba63b BP |
4272 | VLOG_WARN_RL(&rl, "ignoring role request on non-controller " |
4273 | "connection"); | |
4274 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4275 | } | |
4276 | ||
4277 | role = ntohl(nrr->role); | |
4278 | if (role != NX_ROLE_OTHER && role != NX_ROLE_MASTER | |
4279 | && role != NX_ROLE_SLAVE) { | |
4280 | VLOG_WARN_RL(&rl, "received request for unknown role %"PRIu32, role); | |
4281 | ||
4282 | /* There's no good error code for this. */ | |
4283 | return ofp_mkerr(OFPET_BAD_REQUEST, -1); | |
4284 | } | |
4285 | ||
4286 | if (role == NX_ROLE_MASTER) { | |
4287 | struct ofconn *other; | |
4288 | ||
3269c562 | 4289 | HMAP_FOR_EACH (other, hmap_node, &ofconn->ofproto->controllers) { |
9deba63b BP |
4290 | if (other->role == NX_ROLE_MASTER) { |
4291 | other->role = NX_ROLE_SLAVE; | |
4292 | } | |
4293 | } | |
4294 | } | |
4295 | ofconn->role = role; | |
4296 | ||
0bd0c660 BP |
4297 | reply = make_nxmsg_xid(sizeof *reply, NXT_ROLE_REPLY, msg->header.xid, |
4298 | &buf); | |
9deba63b BP |
4299 | reply->role = htonl(role); |
4300 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4301 | ||
4302 | return 0; | |
4303 | } | |
4304 | ||
09246b99 BP |
4305 | static int |
4306 | handle_nxt_set_flow_format(struct ofconn *ofconn, | |
4307 | struct nxt_set_flow_format *msg) | |
4308 | { | |
4309 | uint32_t format; | |
4310 | int error; | |
4311 | ||
4312 | error = check_ofp_message(&msg->header, OFPT_VENDOR, sizeof *msg); | |
4313 | if (error) { | |
4314 | return error; | |
4315 | } | |
4316 | ||
4317 | format = ntohl(msg->format); | |
4318 | if (format == NXFF_OPENFLOW10 | |
4319 | || format == NXFF_TUN_ID_FROM_COOKIE | |
4320 | || format == NXFF_NXM) { | |
4321 | ofconn->flow_format = format; | |
4322 | return 0; | |
4323 | } else { | |
4324 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_EPERM); | |
4325 | } | |
4326 | } | |
4327 | ||
064af421 | 4328 | static int |
3269c562 | 4329 | handle_vendor(struct ofconn *ofconn, void *msg) |
064af421 | 4330 | { |
3269c562 | 4331 | struct ofproto *p = ofconn->ofproto; |
064af421 BP |
4332 | struct ofp_vendor_header *ovh = msg; |
4333 | struct nicira_header *nh; | |
4334 | ||
4335 | if (ntohs(ovh->header.length) < sizeof(struct ofp_vendor_header)) { | |
100e95db | 4336 | VLOG_WARN_RL(&rl, "received vendor message of length %u " |
659586ef JG |
4337 | "(expected at least %zu)", |
4338 | ntohs(ovh->header.length), sizeof(struct ofp_vendor_header)); | |
49bdc010 | 4339 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
4340 | } |
4341 | if (ovh->vendor != htonl(NX_VENDOR_ID)) { | |
4342 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_VENDOR); | |
4343 | } | |
4344 | if (ntohs(ovh->header.length) < sizeof(struct nicira_header)) { | |
100e95db | 4345 | VLOG_WARN_RL(&rl, "received Nicira vendor message of length %u " |
659586ef JG |
4346 | "(expected at least %zu)", |
4347 | ntohs(ovh->header.length), sizeof(struct nicira_header)); | |
49bdc010 | 4348 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_LEN); |
064af421 BP |
4349 | } |
4350 | ||
4351 | nh = msg; | |
4352 | switch (ntohl(nh->subtype)) { | |
4353 | case NXT_STATUS_REQUEST: | |
4354 | return switch_status_handle_request(p->switch_status, ofconn->rconn, | |
4355 | msg); | |
659586ef JG |
4356 | |
4357 | case NXT_TUN_ID_FROM_COOKIE: | |
b70eac89 | 4358 | return handle_tun_id_from_cookie(ofconn, msg); |
9deba63b BP |
4359 | |
4360 | case NXT_ROLE_REQUEST: | |
3269c562 | 4361 | return handle_role_request(ofconn, msg); |
09246b99 BP |
4362 | |
4363 | case NXT_SET_FLOW_FORMAT: | |
4364 | return handle_nxt_set_flow_format(ofconn, msg); | |
4365 | ||
4366 | case NXT_FLOW_MOD: | |
4367 | return handle_nxt_flow_mod(ofconn, &ovh->header); | |
064af421 BP |
4368 | } |
4369 | ||
4370 | return ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_SUBTYPE); | |
4371 | } | |
4372 | ||
246e61ea JP |
4373 | static int |
4374 | handle_barrier_request(struct ofconn *ofconn, struct ofp_header *oh) | |
4375 | { | |
4376 | struct ofp_header *ob; | |
4377 | struct ofpbuf *buf; | |
4378 | ||
4379 | /* Currently, everything executes synchronously, so we can just | |
4380 | * immediately send the barrier reply. */ | |
4381 | ob = make_openflow_xid(sizeof *ob, OFPT_BARRIER_REPLY, oh->xid, &buf); | |
4382 | queue_tx(buf, ofconn, ofconn->reply_counter); | |
4383 | return 0; | |
4384 | } | |
4385 | ||
064af421 | 4386 | static void |
3269c562 | 4387 | handle_openflow(struct ofconn *ofconn, struct ofpbuf *ofp_msg) |
064af421 BP |
4388 | { |
4389 | struct ofp_header *oh = ofp_msg->data; | |
4390 | int error; | |
4391 | ||
4392 | COVERAGE_INC(ofproto_recv_openflow); | |
4393 | switch (oh->type) { | |
4394 | case OFPT_ECHO_REQUEST: | |
4395 | error = handle_echo_request(ofconn, oh); | |
4396 | break; | |
4397 | ||
4398 | case OFPT_ECHO_REPLY: | |
4399 | error = 0; | |
4400 | break; | |
4401 | ||
4402 | case OFPT_FEATURES_REQUEST: | |
3269c562 | 4403 | error = handle_features_request(ofconn, oh); |
064af421 BP |
4404 | break; |
4405 | ||
4406 | case OFPT_GET_CONFIG_REQUEST: | |
3269c562 | 4407 | error = handle_get_config_request(ofconn, oh); |
064af421 BP |
4408 | break; |
4409 | ||
4410 | case OFPT_SET_CONFIG: | |
3269c562 | 4411 | error = handle_set_config(ofconn, ofp_msg->data); |
064af421 BP |
4412 | break; |
4413 | ||
4414 | case OFPT_PACKET_OUT: | |
3269c562 | 4415 | error = handle_packet_out(ofconn, ofp_msg->data); |
064af421 BP |
4416 | break; |
4417 | ||
4418 | case OFPT_PORT_MOD: | |
3269c562 | 4419 | error = handle_port_mod(ofconn, oh); |
064af421 BP |
4420 | break; |
4421 | ||
4422 | case OFPT_FLOW_MOD: | |
09246b99 | 4423 | error = handle_ofpt_flow_mod(ofconn, ofp_msg->data); |
064af421 BP |
4424 | break; |
4425 | ||
4426 | case OFPT_STATS_REQUEST: | |
3269c562 | 4427 | error = handle_stats_request(ofconn, oh); |
064af421 BP |
4428 | break; |
4429 | ||
4430 | case OFPT_VENDOR: | |
3269c562 | 4431 | error = handle_vendor(ofconn, ofp_msg->data); |
064af421 BP |
4432 | break; |
4433 | ||
246e61ea JP |
4434 | case OFPT_BARRIER_REQUEST: |
4435 | error = handle_barrier_request(ofconn, oh); | |
4436 | break; | |
4437 | ||
064af421 BP |
4438 | default: |
4439 | if (VLOG_IS_WARN_ENABLED()) { | |
4440 | char *s = ofp_to_string(oh, ntohs(oh->length), 2); | |
4441 | VLOG_DBG_RL(&rl, "OpenFlow message ignored: %s", s); | |
4442 | free(s); | |
4443 | } | |
4444 | error = ofp_mkerr(OFPET_BAD_REQUEST, OFPBRC_BAD_TYPE); | |
4445 | break; | |
4446 | } | |
4447 | ||
4448 | if (error) { | |
4449 | send_error_oh(ofconn, ofp_msg->data, error); | |
4450 | } | |
4451 | } | |
4452 | \f | |
4453 | static void | |
72b06300 | 4454 | handle_odp_miss_msg(struct ofproto *p, struct ofpbuf *packet) |
064af421 BP |
4455 | { |
4456 | struct odp_msg *msg = packet->data; | |
064af421 | 4457 | struct ofpbuf payload; |
bcf84111 | 4458 | struct facet *facet; |
ae412e7d | 4459 | struct flow flow; |
064af421 | 4460 | |
064af421 BP |
4461 | payload.data = msg + 1; |
4462 | payload.size = msg->length - sizeof *msg; | |
659586ef | 4463 | flow_extract(&payload, msg->arg, msg->port, &flow); |
064af421 | 4464 | |
4617e2c1 EJ |
4465 | packet->l2 = payload.l2; |
4466 | packet->l3 = payload.l3; | |
4467 | packet->l4 = payload.l4; | |
4468 | packet->l7 = payload.l7; | |
4469 | ||
0ad9b732 JP |
4470 | /* Check with in-band control to see if this packet should be sent |
4471 | * to the local port regardless of the flow table. */ | |
4472 | if (in_band_msg_in_hook(p->in_band, &flow, &payload)) { | |
4473 | union odp_action action; | |
4474 | ||
4475 | memset(&action, 0, sizeof(action)); | |
4476 | action.output.type = ODPAT_OUTPUT; | |
4477 | action.output.port = ODPP_LOCAL; | |
f1588b1f | 4478 | dpif_execute(p->dpif, &action, 1, &payload); |
0ad9b732 JP |
4479 | } |
4480 | ||
bcf84111 BP |
4481 | facet = facet_lookup_valid(p, &flow); |
4482 | if (!facet) { | |
4483 | struct rule *rule = rule_lookup(p, &flow); | |
4484 | if (!rule) { | |
4485 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
4486 | struct ofport *port = get_port(p, msg->port); | |
4487 | if (port) { | |
4488 | if (port->opp.config & OFPPC_NO_PACKET_IN) { | |
4489 | COVERAGE_INC(ofproto_no_packet_in); | |
4490 | /* XXX install 'drop' flow entry */ | |
4491 | ofpbuf_delete(packet); | |
4492 | return; | |
4493 | } | |
4494 | } else { | |
4495 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
4496 | msg->port); | |
064af421 | 4497 | } |
064af421 | 4498 | |
bcf84111 BP |
4499 | COVERAGE_INC(ofproto_packet_in); |
4500 | send_packet_in(p, packet); | |
4501 | return; | |
064af421 | 4502 | } |
bcf84111 BP |
4503 | |
4504 | facet = facet_create(p, rule, &flow, packet); | |
4505 | } else if (!facet->may_install) { | |
4506 | /* The facet is not installable, that is, we need to process every | |
4507 | * packet, so process the current packet's actions into 'facet'. */ | |
4508 | facet_make_actions(p, facet, packet); | |
064af421 BP |
4509 | } |
4510 | ||
bcf84111 | 4511 | if (facet->rule->cr.priority == FAIL_OPEN_PRIORITY) { |
7778bd15 BP |
4512 | /* |
4513 | * Extra-special case for fail-open mode. | |
4514 | * | |
4515 | * We are in fail-open mode and the packet matched the fail-open rule, | |
4516 | * but we are connected to a controller too. We should send the packet | |
4517 | * up to the controller in the hope that it will try to set up a flow | |
4518 | * and thereby allow us to exit fail-open. | |
4519 | * | |
4520 | * See the top-level comment in fail-open.c for more information. | |
4521 | */ | |
c9b5816c BP |
4522 | send_packet_in(p, ofpbuf_clone_with_headroom(packet, |
4523 | DPIF_RECV_MSG_PADDING)); | |
7778bd15 | 4524 | } |
750638bb BP |
4525 | |
4526 | ofpbuf_pull(packet, sizeof *msg); | |
bcf84111 BP |
4527 | facet_execute(p, facet, packet); |
4528 | facet_install(p, facet, false); | |
064af421 | 4529 | } |
72b06300 BP |
4530 | |
4531 | static void | |
4532 | handle_odp_msg(struct ofproto *p, struct ofpbuf *packet) | |
4533 | { | |
4534 | struct odp_msg *msg = packet->data; | |
4535 | ||
4536 | switch (msg->type) { | |
4537 | case _ODPL_ACTION_NR: | |
4538 | COVERAGE_INC(ofproto_ctlr_action); | |
76ce9432 | 4539 | send_packet_in(p, packet); |
72b06300 BP |
4540 | break; |
4541 | ||
4542 | case _ODPL_SFLOW_NR: | |
4543 | if (p->sflow) { | |
4544 | ofproto_sflow_received(p->sflow, msg); | |
4545 | } | |
4546 | ofpbuf_delete(packet); | |
4547 | break; | |
4548 | ||
4549 | case _ODPL_MISS_NR: | |
4550 | handle_odp_miss_msg(p, packet); | |
4551 | break; | |
4552 | ||
4553 | default: | |
4554 | VLOG_WARN_RL(&rl, "received ODP message of unexpected type %"PRIu32, | |
4555 | msg->type); | |
4556 | break; | |
4557 | } | |
4558 | } | |
064af421 | 4559 | \f |
4a4cdb3b BP |
4560 | /* Flow expiration. */ |
4561 | ||
0de7a4b4 | 4562 | static int ofproto_dp_max_idle(const struct ofproto *); |
4a4cdb3b | 4563 | static void ofproto_update_used(struct ofproto *); |
5ecc9d81 | 4564 | static void rule_expire(struct ofproto *, struct rule *); |
bcf84111 | 4565 | static void ofproto_expire_facets(struct ofproto *, int dp_max_idle); |
4a4cdb3b BP |
4566 | |
4567 | /* This function is called periodically by ofproto_run(). Its job is to | |
4568 | * collect updates for the flows that have been installed into the datapath, | |
4569 | * most importantly when they last were used, and then use that information to | |
0de7a4b4 BP |
4570 | * expire flows that have not been used recently. |
4571 | * | |
4572 | * Returns the number of milliseconds after which it should be called again. */ | |
4573 | static int | |
4a4cdb3b BP |
4574 | ofproto_expire(struct ofproto *ofproto) |
4575 | { | |
5ecc9d81 BP |
4576 | struct rule *rule, *next_rule; |
4577 | struct cls_cursor cursor; | |
4578 | int dp_max_idle; | |
4a4cdb3b BP |
4579 | |
4580 | /* Update 'used' for each flow in the datapath. */ | |
4581 | ofproto_update_used(ofproto); | |
4582 | ||
bcf84111 | 4583 | /* Expire facets that have been idle too long. */ |
5ecc9d81 BP |
4584 | dp_max_idle = ofproto_dp_max_idle(ofproto); |
4585 | ofproto_expire_facets(ofproto, dp_max_idle); | |
bcf84111 BP |
4586 | |
4587 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
5ecc9d81 BP |
4588 | cls_cursor_init(&cursor, &ofproto->cls, NULL); |
4589 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) { | |
4590 | rule_expire(ofproto, rule); | |
4591 | } | |
4a4cdb3b BP |
4592 | |
4593 | /* Let the hook know that we're at a stable point: all outstanding data | |
4594 | * in existing flows has been accounted to the account_cb. Thus, the | |
4595 | * hook can now reasonably do operations that depend on having accurate | |
4596 | * flow volume accounting (currently, that's just bond rebalancing). */ | |
4597 | if (ofproto->ofhooks->account_checkpoint_cb) { | |
4598 | ofproto->ofhooks->account_checkpoint_cb(ofproto->aux); | |
4599 | } | |
0de7a4b4 | 4600 | |
5ecc9d81 | 4601 | return MIN(dp_max_idle, 1000); |
4a4cdb3b BP |
4602 | } |
4603 | ||
bcf84111 | 4604 | /* Update 'used' member of installed facets. */ |
4a4cdb3b BP |
4605 | static void |
4606 | ofproto_update_used(struct ofproto *p) | |
4607 | { | |
4608 | struct odp_flow *flows; | |
4609 | size_t n_flows; | |
4610 | size_t i; | |
4611 | int error; | |
4612 | ||
4613 | error = dpif_flow_list_all(p->dpif, &flows, &n_flows); | |
4614 | if (error) { | |
4615 | return; | |
4616 | } | |
4617 | ||
4618 | for (i = 0; i < n_flows; i++) { | |
4619 | struct odp_flow *f = &flows[i]; | |
bcf84111 | 4620 | struct facet *facet; |
ae412e7d | 4621 | struct flow flow; |
14608a15 BP |
4622 | |
4623 | odp_flow_key_to_flow(&f->key, &flow); | |
bcf84111 | 4624 | facet = facet_find(p, &flow); |
4a4cdb3b | 4625 | |
bcf84111 BP |
4626 | if (facet && facet->installed) { |
4627 | facet_update_time(p, facet, &f->stats); | |
4628 | facet_account(p, facet, f->stats.n_bytes); | |
4a4cdb3b BP |
4629 | } else { |
4630 | /* There's a flow in the datapath that we know nothing about. | |
4631 | * Delete it. */ | |
4632 | COVERAGE_INC(ofproto_unexpected_rule); | |
4633 | dpif_flow_del(p->dpif, f); | |
4634 | } | |
4635 | ||
4636 | } | |
4637 | free(flows); | |
4638 | } | |
4639 | ||
0de7a4b4 | 4640 | /* Calculates and returns the number of milliseconds of idle time after which |
bcf84111 | 4641 | * facets should expire from the datapath and we should fold their statistics |
0de7a4b4 BP |
4642 | * into their parent rules in userspace. */ |
4643 | static int | |
4644 | ofproto_dp_max_idle(const struct ofproto *ofproto) | |
4645 | { | |
4646 | /* | |
4647 | * Idle time histogram. | |
4648 | * | |
bcf84111 | 4649 | * Most of the time a switch has a relatively small number of facets. When |
0de7a4b4 BP |
4650 | * this is the case we might as well keep statistics for all of them in |
4651 | * userspace and to cache them in the kernel datapath for performance as | |
4652 | * well. | |
4653 | * | |
bcf84111 | 4654 | * As the number of facets increases, the memory required to maintain |
0de7a4b4 | 4655 | * statistics about them in userspace and in the kernel becomes |
bcf84111 | 4656 | * significant. However, with a large number of facets it is likely that |
0de7a4b4 BP |
4657 | * only a few of them are "heavy hitters" that consume a large amount of |
4658 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
bcf84111 | 4659 | * kernel and maintaining in userspaces; other facets we can discard. |
0de7a4b4 BP |
4660 | * |
4661 | * The technique used to compute the idle time is to build a histogram with | |
bcf84111 BP |
4662 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet |
4663 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
0de7a4b4 | 4664 | * After the histogram has been built, we compute the cutoff so that only |
bcf84111 BP |
4665 | * the most-recently-used 1% of facets (but at least 1000 flows) are kept |
4666 | * cached. At least the most-recently-used bucket of facets is kept, so | |
4667 | * actually an arbitrary number of facets can be kept in any given | |
0de7a4b4 BP |
4668 | * expiration run (though the next run will delete most of those unless |
4669 | * they receive additional data). | |
4670 | * | |
bcf84111 BP |
4671 | * This requires a second pass through the facets, in addition to the pass |
4672 | * made by ofproto_update_used(), because the former function never looks | |
4673 | * at uninstallable facets. | |
0de7a4b4 BP |
4674 | */ |
4675 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
4676 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
4677 | int buckets[N_BUCKETS] = { 0 }; | |
bcf84111 | 4678 | struct facet *facet; |
0de7a4b4 | 4679 | int total, bucket; |
0de7a4b4 BP |
4680 | long long int now; |
4681 | int i; | |
4682 | ||
bcf84111 | 4683 | total = hmap_count(&ofproto->facets); |
0de7a4b4 BP |
4684 | if (total <= 1000) { |
4685 | return N_BUCKETS * BUCKET_WIDTH; | |
4686 | } | |
4687 | ||
4688 | /* Build histogram. */ | |
4689 | now = time_msec(); | |
bcf84111 BP |
4690 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { |
4691 | long long int idle = now - facet->used; | |
0de7a4b4 BP |
4692 | int bucket = (idle <= 0 ? 0 |
4693 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
4694 | : (unsigned int) idle / BUCKET_WIDTH); | |
4695 | buckets[bucket]++; | |
4696 | } | |
4697 | ||
4698 | /* Find the first bucket whose flows should be expired. */ | |
4699 | for (bucket = 0; bucket < N_BUCKETS; bucket++) { | |
4700 | if (buckets[bucket]) { | |
4701 | int subtotal = 0; | |
4702 | do { | |
4703 | subtotal += buckets[bucket++]; | |
4704 | } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); | |
4705 | break; | |
4706 | } | |
4707 | } | |
4708 | ||
4709 | if (VLOG_IS_DBG_ENABLED()) { | |
4710 | struct ds s; | |
4711 | ||
4712 | ds_init(&s); | |
4713 | ds_put_cstr(&s, "keep"); | |
4714 | for (i = 0; i < N_BUCKETS; i++) { | |
4715 | if (i == bucket) { | |
4716 | ds_put_cstr(&s, ", drop"); | |
4717 | } | |
4718 | if (buckets[i]) { | |
4719 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
4720 | } | |
4721 | } | |
4722 | VLOG_INFO("%s: %s (msec:count)", | |
4723 | dpif_name(ofproto->dpif), ds_cstr(&s)); | |
4724 | ds_destroy(&s); | |
4725 | } | |
4726 | ||
4727 | return bucket * BUCKET_WIDTH; | |
4728 | } | |
4729 | ||
4a4cdb3b | 4730 | static void |
bcf84111 | 4731 | facet_active_timeout(struct ofproto *ofproto, struct facet *facet) |
4a4cdb3b | 4732 | { |
bcf84111 BP |
4733 | if (ofproto->netflow && !facet_is_controller_flow(facet) && |
4734 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
4a4cdb3b BP |
4735 | struct ofexpired expired; |
4736 | struct odp_flow odp_flow; | |
4737 | ||
4738 | /* Get updated flow stats. | |
4739 | * | |
4740 | * XXX We could avoid this call entirely if (1) ofproto_update_used() | |
4741 | * updated TCP flags and (2) the dpif_flow_list_all() in | |
4742 | * ofproto_update_used() zeroed TCP flags. */ | |
4743 | memset(&odp_flow, 0, sizeof odp_flow); | |
bcf84111 BP |
4744 | if (facet->installed) { |
4745 | odp_flow_key_from_flow(&odp_flow.key, &facet->flow); | |
4a4cdb3b BP |
4746 | odp_flow.flags = ODPFF_ZERO_TCP_FLAGS; |
4747 | dpif_flow_get(ofproto->dpif, &odp_flow); | |
4748 | ||
4749 | if (odp_flow.stats.n_packets) { | |
bcf84111 BP |
4750 | facet_update_time(ofproto, facet, &odp_flow.stats); |
4751 | netflow_flow_update_flags(&facet->nf_flow, | |
4a4cdb3b BP |
4752 | odp_flow.stats.tcp_flags); |
4753 | } | |
4754 | } | |
4755 | ||
bcf84111 BP |
4756 | expired.flow = facet->flow; |
4757 | expired.packet_count = facet->packet_count + | |
4a4cdb3b | 4758 | odp_flow.stats.n_packets; |
bcf84111 BP |
4759 | expired.byte_count = facet->byte_count + odp_flow.stats.n_bytes; |
4760 | expired.used = facet->used; | |
4761 | ||
4762 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
4763 | } | |
4764 | } | |
4a4cdb3b | 4765 | |
bcf84111 BP |
4766 | static void |
4767 | ofproto_expire_facets(struct ofproto *ofproto, int dp_max_idle) | |
4768 | { | |
4769 | long long int cutoff = time_msec() - dp_max_idle; | |
4770 | struct facet *facet, *next_facet; | |
4771 | ||
4772 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
4773 | facet_active_timeout(ofproto, facet); | |
4774 | if (facet->used < cutoff) { | |
4775 | facet_remove(ofproto, facet); | |
4776 | } | |
4a4cdb3b BP |
4777 | } |
4778 | } | |
4779 | ||
5ecc9d81 BP |
4780 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, |
4781 | * then delete it entirely. */ | |
4a4cdb3b | 4782 | static void |
5ecc9d81 | 4783 | rule_expire(struct ofproto *ofproto, struct rule *rule) |
4a4cdb3b | 4784 | { |
bcf84111 BP |
4785 | struct facet *facet, *next_facet; |
4786 | long long int now; | |
4787 | uint8_t reason; | |
4a4cdb3b | 4788 | |
bcf84111 | 4789 | /* Has 'rule' expired? */ |
4a4cdb3b | 4790 | now = time_msec(); |
bcf84111 BP |
4791 | if (rule->hard_timeout |
4792 | && now > rule->created + rule->hard_timeout * 1000) { | |
4793 | reason = OFPRR_HARD_TIMEOUT; | |
4794 | } else if (rule->idle_timeout && list_is_empty(&rule->facets) | |
4795 | && now >rule->used + rule->idle_timeout * 1000) { | |
4796 | reason = OFPRR_IDLE_TIMEOUT; | |
4a4cdb3b | 4797 | } else { |
bcf84111 | 4798 | return; |
4a4cdb3b | 4799 | } |
064af421 | 4800 | |
bcf84111 | 4801 | COVERAGE_INC(ofproto_expired); |
064af421 | 4802 | |
bcf84111 BP |
4803 | /* Update stats. (This is a no-op if the rule expired due to an idle |
4804 | * timeout, because that only happens when the rule has no facets left.) */ | |
4805 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
5ecc9d81 | 4806 | facet_remove(ofproto, facet); |
064af421 BP |
4807 | } |
4808 | ||
bcf84111 BP |
4809 | /* Get rid of the rule. */ |
4810 | if (!rule_is_hidden(rule)) { | |
5ecc9d81 | 4811 | rule_send_removed(ofproto, rule, reason); |
bcf84111 | 4812 | } |
5ecc9d81 | 4813 | rule_remove(ofproto, rule); |
064af421 | 4814 | } |
bcf84111 | 4815 | \f |
064af421 | 4816 | static struct ofpbuf * |
09246b99 BP |
4817 | compose_ofp_flow_removed(struct ofconn *ofconn, const struct rule *rule, |
4818 | uint8_t reason) | |
064af421 | 4819 | { |
ca069229 | 4820 | struct ofp_flow_removed *ofr; |
064af421 BP |
4821 | struct ofpbuf *buf; |
4822 | ||
ca069229 | 4823 | ofr = make_openflow(sizeof *ofr, OFPT_FLOW_REMOVED, &buf); |
d8ae4d67 | 4824 | ofputil_cls_rule_to_match(&rule->cr, ofconn->flow_format, &ofr->match); |
39997502 | 4825 | ofr->cookie = rule->flow_cookie; |
ca069229 JP |
4826 | ofr->priority = htons(rule->cr.priority); |
4827 | ofr->reason = reason; | |
c6ebb8fb | 4828 | calc_flow_duration(rule->created, &ofr->duration_sec, &ofr->duration_nsec); |
ca069229 JP |
4829 | ofr->idle_timeout = htons(rule->idle_timeout); |
4830 | ofr->packet_count = htonll(rule->packet_count); | |
4831 | ofr->byte_count = htonll(rule->byte_count); | |
064af421 BP |
4832 | |
4833 | return buf; | |
4834 | } | |
4835 | ||
09246b99 BP |
4836 | static struct ofpbuf * |
4837 | compose_nx_flow_removed(const struct rule *rule, uint8_t reason) | |
4838 | { | |
4839 | struct nx_flow_removed *nfr; | |
4840 | struct ofpbuf *buf; | |
4841 | int match_len; | |
4842 | ||
4843 | nfr = make_nxmsg(sizeof *nfr, NXT_FLOW_REMOVED, &buf); | |
4844 | ||
4845 | match_len = nx_put_match(buf, &rule->cr); | |
4846 | ||
4847 | nfr->cookie = rule->flow_cookie; | |
4848 | nfr->priority = htons(rule->cr.priority); | |
4849 | nfr->reason = reason; | |
4850 | calc_flow_duration(rule->created, &nfr->duration_sec, &nfr->duration_nsec); | |
4851 | nfr->idle_timeout = htons(rule->idle_timeout); | |
4852 | nfr->match_len = htons(match_len); | |
4853 | nfr->packet_count = htonll(rule->packet_count); | |
4854 | nfr->byte_count = htonll(rule->byte_count); | |
4855 | ||
4856 | return buf; | |
4857 | } | |
4858 | ||
ca069229 | 4859 | static void |
bcf84111 | 4860 | rule_send_removed(struct ofproto *p, struct rule *rule, uint8_t reason) |
064af421 BP |
4861 | { |
4862 | struct ofconn *ofconn; | |
064af421 | 4863 | |
3b587616 BP |
4864 | if (!rule->send_flow_removed) { |
4865 | return; | |
4866 | } | |
4867 | ||
4e8e4213 | 4868 | LIST_FOR_EACH (ofconn, node, &p->all_conns) { |
7a0efeb5 BP |
4869 | struct ofpbuf *msg; |
4870 | ||
4871 | if (!rconn_is_connected(ofconn->rconn) | |
4872 | || !ofconn_receives_async_msgs(ofconn)) { | |
4873 | continue; | |
064af421 | 4874 | } |
7a0efeb5 | 4875 | |
09246b99 BP |
4876 | msg = (ofconn->flow_format == NXFF_NXM |
4877 | ? compose_nx_flow_removed(rule, reason) | |
4878 | : compose_ofp_flow_removed(ofconn, rule, reason)); | |
6d6c7259 BP |
4879 | |
4880 | /* Account flow expirations under ofconn->reply_counter, the counter | |
4881 | * for replies to OpenFlow requests. That works because preventing | |
4882 | * OpenFlow requests from being processed also prevents new flows from | |
4883 | * being added (and expiring). (It also prevents processing OpenFlow | |
4884 | * requests that would not add new flows, so it is imperfect.) */ | |
7a0efeb5 | 4885 | queue_tx(msg, ofconn, ofconn->reply_counter); |
064af421 BP |
4886 | } |
4887 | } | |
4888 | ||
43253595 | 4889 | /* pinsched callback for sending 'packet' on 'ofconn'. */ |
064af421 | 4890 | static void |
76ce9432 | 4891 | do_send_packet_in(struct ofpbuf *packet, void *ofconn_) |
064af421 | 4892 | { |
76ce9432 | 4893 | struct ofconn *ofconn = ofconn_; |
43253595 BP |
4894 | |
4895 | rconn_send_with_limit(ofconn->rconn, packet, | |
4896 | ofconn->packet_in_counter, 100); | |
4897 | } | |
4898 | ||
4899 | /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and | |
4900 | * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s | |
4901 | * packet scheduler for sending. | |
4902 | * | |
30ea5d93 BP |
4903 | * 'max_len' specifies the maximum number of bytes of the packet to send on |
4904 | * 'ofconn' (INT_MAX specifies no limit). | |
4905 | * | |
43253595 BP |
4906 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, |
4907 | * ownership is transferred to this function. */ | |
4908 | static void | |
30ea5d93 BP |
4909 | schedule_packet_in(struct ofconn *ofconn, struct ofpbuf *packet, int max_len, |
4910 | bool clone) | |
43253595 | 4911 | { |
76ce9432 | 4912 | struct ofproto *ofproto = ofconn->ofproto; |
43253595 BP |
4913 | struct ofp_packet_in *opi = packet->data; |
4914 | uint16_t in_port = ofp_port_to_odp_port(ntohs(opi->in_port)); | |
4915 | int send_len, trim_size; | |
76ce9432 | 4916 | uint32_t buffer_id; |
064af421 | 4917 | |
43253595 BP |
4918 | /* Get buffer. */ |
4919 | if (opi->reason == OFPR_ACTION) { | |
76ce9432 | 4920 | buffer_id = UINT32_MAX; |
43253595 BP |
4921 | } else if (ofproto->fail_open && fail_open_is_active(ofproto->fail_open)) { |
4922 | buffer_id = pktbuf_get_null(); | |
89b9612d BP |
4923 | } else if (!ofconn->pktbuf) { |
4924 | buffer_id = UINT32_MAX; | |
76ce9432 | 4925 | } else { |
43253595 BP |
4926 | struct ofpbuf payload; |
4927 | payload.data = opi->data; | |
4928 | payload.size = packet->size - offsetof(struct ofp_packet_in, data); | |
4929 | buffer_id = pktbuf_save(ofconn->pktbuf, &payload, in_port); | |
76ce9432 | 4930 | } |
372179d4 | 4931 | |
43253595 BP |
4932 | /* Figure out how much of the packet to send. */ |
4933 | send_len = ntohs(opi->total_len); | |
4934 | if (buffer_id != UINT32_MAX) { | |
4935 | send_len = MIN(send_len, ofconn->miss_send_len); | |
4936 | } | |
30ea5d93 | 4937 | send_len = MIN(send_len, max_len); |
064af421 | 4938 | |
43253595 BP |
4939 | /* Adjust packet length and clone if necessary. */ |
4940 | trim_size = offsetof(struct ofp_packet_in, data) + send_len; | |
4941 | if (clone) { | |
4942 | packet = ofpbuf_clone_data(packet->data, trim_size); | |
4943 | opi = packet->data; | |
4944 | } else { | |
4945 | packet->size = trim_size; | |
4946 | } | |
4947 | ||
4948 | /* Update packet headers. */ | |
4949 | opi->buffer_id = htonl(buffer_id); | |
4950 | update_openflow_length(packet); | |
4951 | ||
4952 | /* Hand over to packet scheduler. It might immediately call into | |
4953 | * do_send_packet_in() or it might buffer it for a while (until a later | |
4954 | * call to pinsched_run()). */ | |
4955 | pinsched_send(ofconn->schedulers[opi->reason], in_port, | |
4956 | packet, do_send_packet_in, ofconn); | |
064af421 BP |
4957 | } |
4958 | ||
43253595 BP |
4959 | /* Replace struct odp_msg header in 'packet' by equivalent struct |
4960 | * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as | |
4961 | * returned by dpif_recv()). | |
4962 | * | |
4963 | * The conversion is not complete: the caller still needs to trim any unneeded | |
4964 | * payload off the end of the buffer, set the length in the OpenFlow header, | |
4965 | * and set buffer_id. Those require us to know the controller settings and so | |
30ea5d93 BP |
4966 | * must be done on a per-controller basis. |
4967 | * | |
4968 | * Returns the maximum number of bytes of the packet that should be sent to | |
4969 | * the controller (INT_MAX if no limit). */ | |
4970 | static int | |
43253595 | 4971 | do_convert_to_packet_in(struct ofpbuf *packet) |
064af421 | 4972 | { |
76ce9432 | 4973 | struct odp_msg *msg = packet->data; |
43253595 BP |
4974 | struct ofp_packet_in *opi; |
4975 | uint8_t reason; | |
4976 | uint16_t total_len; | |
4977 | uint16_t in_port; | |
30ea5d93 | 4978 | int max_len; |
43253595 BP |
4979 | |
4980 | /* Extract relevant header fields */ | |
30ea5d93 BP |
4981 | if (msg->type == _ODPL_ACTION_NR) { |
4982 | reason = OFPR_ACTION; | |
4983 | max_len = msg->arg; | |
4984 | } else { | |
4985 | reason = OFPR_NO_MATCH; | |
4986 | max_len = INT_MAX; | |
4987 | } | |
43253595 BP |
4988 | total_len = msg->length - sizeof *msg; |
4989 | in_port = odp_port_to_ofp_port(msg->port); | |
4990 | ||
4991 | /* Repurpose packet buffer by overwriting header. */ | |
4992 | ofpbuf_pull(packet, sizeof(struct odp_msg)); | |
4993 | opi = ofpbuf_push_zeros(packet, offsetof(struct ofp_packet_in, data)); | |
4994 | opi->header.version = OFP_VERSION; | |
4995 | opi->header.type = OFPT_PACKET_IN; | |
4996 | opi->total_len = htons(total_len); | |
4997 | opi->in_port = htons(in_port); | |
4998 | opi->reason = reason; | |
30ea5d93 BP |
4999 | |
5000 | return max_len; | |
43253595 BP |
5001 | } |
5002 | ||
5003 | /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or | |
5004 | * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller | |
5005 | * as necessary according to their individual configurations. | |
5006 | * | |
5007 | * 'packet' must have sufficient headroom to convert it into a struct | |
5008 | * ofp_packet_in (e.g. as returned by dpif_recv()). | |
5009 | * | |
5010 | * Takes ownership of 'packet'. */ | |
5011 | static void | |
5012 | send_packet_in(struct ofproto *ofproto, struct ofpbuf *packet) | |
5013 | { | |
76ce9432 | 5014 | struct ofconn *ofconn, *prev; |
30ea5d93 | 5015 | int max_len; |
064af421 | 5016 | |
30ea5d93 | 5017 | max_len = do_convert_to_packet_in(packet); |
76ce9432 BP |
5018 | |
5019 | prev = NULL; | |
4e8e4213 | 5020 | LIST_FOR_EACH (ofconn, node, &ofproto->all_conns) { |
c91248b3 | 5021 | if (ofconn_receives_async_msgs(ofconn)) { |
9deba63b | 5022 | if (prev) { |
30ea5d93 | 5023 | schedule_packet_in(prev, packet, max_len, true); |
9deba63b BP |
5024 | } |
5025 | prev = ofconn; | |
064af421 | 5026 | } |
76ce9432 BP |
5027 | } |
5028 | if (prev) { | |
30ea5d93 | 5029 | schedule_packet_in(prev, packet, max_len, false); |
76ce9432 BP |
5030 | } else { |
5031 | ofpbuf_delete(packet); | |
064af421 | 5032 | } |
064af421 BP |
5033 | } |
5034 | ||
5035 | static uint64_t | |
fa60c019 | 5036 | pick_datapath_id(const struct ofproto *ofproto) |
064af421 | 5037 | { |
fa60c019 | 5038 | const struct ofport *port; |
064af421 | 5039 | |
ca0f572c | 5040 | port = get_port(ofproto, ODPP_LOCAL); |
fa60c019 BP |
5041 | if (port) { |
5042 | uint8_t ea[ETH_ADDR_LEN]; | |
5043 | int error; | |
5044 | ||
5045 | error = netdev_get_etheraddr(port->netdev, ea); | |
064af421 BP |
5046 | if (!error) { |
5047 | return eth_addr_to_uint64(ea); | |
5048 | } | |
5049 | VLOG_WARN("could not get MAC address for %s (%s)", | |
fa60c019 | 5050 | netdev_get_name(port->netdev), strerror(error)); |
064af421 | 5051 | } |
fa60c019 | 5052 | return ofproto->fallback_dpid; |
064af421 BP |
5053 | } |
5054 | ||
5055 | static uint64_t | |
5056 | pick_fallback_dpid(void) | |
5057 | { | |
5058 | uint8_t ea[ETH_ADDR_LEN]; | |
70150daf | 5059 | eth_addr_nicira_random(ea); |
064af421 BP |
5060 | return eth_addr_to_uint64(ea); |
5061 | } | |
5062 | \f | |
5063 | static bool | |
ae412e7d | 5064 | default_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, |
064af421 | 5065 | struct odp_actions *actions, tag_type *tags, |
6a07af36 | 5066 | uint16_t *nf_output_iface, void *ofproto_) |
064af421 BP |
5067 | { |
5068 | struct ofproto *ofproto = ofproto_; | |
5069 | int out_port; | |
5070 | ||
5071 | /* Drop frames for reserved multicast addresses. */ | |
5072 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
5073 | return true; | |
5074 | } | |
5075 | ||
5076 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
5077 | if (packet != NULL) { | |
5078 | tag_type rev_tag = mac_learning_learn(ofproto->ml, flow->dl_src, | |
7febb910 JG |
5079 | 0, flow->in_port, |
5080 | GRAT_ARP_LOCK_NONE); | |
064af421 BP |
5081 | if (rev_tag) { |
5082 | /* The log messages here could actually be useful in debugging, | |
5083 | * so keep the rate limit relatively high. */ | |
5084 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
5085 | VLOG_DBG_RL(&rl, "learned that "ETH_ADDR_FMT" is on port %"PRIu16, | |
5086 | ETH_ADDR_ARGS(flow->dl_src), flow->in_port); | |
5087 | ofproto_revalidate(ofproto, rev_tag); | |
5088 | } | |
5089 | } | |
5090 | ||
5091 | /* Determine output port. */ | |
7febb910 JG |
5092 | out_port = mac_learning_lookup_tag(ofproto->ml, flow->dl_dst, 0, tags, |
5093 | NULL); | |
064af421 | 5094 | if (out_port < 0) { |
f1588b1f BP |
5095 | flood_packets(ofproto, flow->in_port, OFPPC_NO_FLOOD, |
5096 | nf_output_iface, actions); | |
064af421 BP |
5097 | } else if (out_port != flow->in_port) { |
5098 | odp_actions_add(actions, ODPAT_OUTPUT)->output.port = out_port; | |
6a07af36 | 5099 | *nf_output_iface = out_port; |
064af421 BP |
5100 | } else { |
5101 | /* Drop. */ | |
5102 | } | |
5103 | ||
5104 | return true; | |
5105 | } | |
5106 | ||
5107 | static const struct ofhooks default_ofhooks = { | |
064af421 BP |
5108 | default_normal_ofhook_cb, |
5109 | NULL, | |
5110 | NULL | |
5111 | }; |